Sign Up
Login
Try for Free
Login
The GARS-3 (Gilliam Autism Rating Scale, Third Edition) is an informant rating scale for ages 3 through 22: a parent, teacher, or clinician rates items across six subscales, producing an Autism Index and probability language. Schools and autism teams use it as supporting evidence, never as a diagnosis. The write-up must name the respondent and index used. This page covers how to write up GARS-3 results, with a fictional sample.
School psychologists, educational diagnosticians, psychologists, and clinicians scoring parent or teacher ratings; PRO-ED test level B (WPS lists C)
IEP and autism evaluation teams, diagnosing clinicians, ABA reviewers and payers, referring physicians, families
250 to 600 words for the results section · the respondent completes the rating in about 5 to 10 minutes; scoring and integration add clinician time
Norm-referenced informant rating scale (six DSM-5-aligned subscales; Autism Index 4 or 6, mean 100, SD 15)
School autism evaluations and re-evaluations, clinic batteries as standardized informant evidence, ages 3 through 22, alongside observation and interview measures
Published by PRO-ED (©2014; no fourth edition as of August 2026); scored by manual or PRO-ED's online system; no items, anchors, or cut values reproduced
The GARS-3 (Gilliam Autism Rating Scale, Third Edition; James E. Gilliam; PRO-ED, ©2014, with some distributor listings dating the release 2013) is a norm-referenced informant rating scale for ages 3 through 22. A parent, teacher, or clinician who knows the examinee rates 56 items (current publisher pages say 56; some peer-reviewed papers report 58, a discrepancy worth checking against the licensed form) on a four-point frequency scale, across six DSM-5-aligned subscales: Restrictive/Repetitive Behaviors, Social Interaction, Social Communication, Emotional Responses, Cognitive Style, and Maladaptive Speech. Subscale raw scores convert to scaled scores (mean 10, SD 3) and combine into one of two composites: Autism Index 4, from the first four subscales, for individuals whose speech is absent or severely limited, or Autism Index 6, from all six, for individuals with sufficient functional speech. The index (mean 100, SD 15) maps to the manual's ordered probability-of-autism classifications and to a severity framing the publisher aligns with DSM-5 support levels. PRO-ED is the publisher; WPS, Pearson, and PAR distribute it (Pearson's own documentation says it distributes but does not publish), and a Spanish version, the GARS-3S, has its own forms and manual.
The write-up's center of gravity is what this instrument is: one informant's standardized report. The rating quantifies how closely that respondent's account resembles the behavioral pattern in the instrument's norms, which were drawn entirely from individuals already diagnosed with autism, a fact a 2017 published review highlighted and almost no explainer page mentions. So the score is resemblance to an autism reference sample, filtered through one rater in one setting: evidence to integrate, never a verdict. The probability language is a classification label, not a literal percentage, and neither a diagnosis nor IDEA eligibility nor an insurance authorization follows from it alone. The instruments the GARS-3 must converge with have their own pages: structured observation on the ADOS-2 page, the caregiver developmental interview on the ADI-R page, and the full synthesis on the autism evaluation report page.
School teams are the heaviest users: the scale is brief (5 to 10 minutes for the respondent), inexpensive, and built for teacher and parent report, so it appears constantly in IDEA autism evaluations and re-evaluations, where federal law requires a variety of tools and prohibits any single measure from deciding eligibility. Clinics use it as the standardized informant component of diagnostic batteries, alongside clinician ratings covered on the CARS-2 page, trait measures covered on the SRS-2 page, and the observation and interview instruments payers actually name. Payer reviewers meet it inside psychological-testing claims (the 96130 to 96139 code family) and diagnostic packets for ABA authorization, where its role is strictly supporting: TRICARE's autism program lists the GARS among accepted instruments while stating a parent questionnaire alone is insufficient, and North Carolina Medicaid's policy amended August 2026 names other procedures entirely for a qualifying diagnosis. The report's readers are the IEP team applying the state's educational definition, the diagnosing clinician weighing convergence, the reviewer checking whether the packet contains more than a rating scale, and a family who was told a number and deserves to know what it does and does not mean. Toddler-age screening that precedes all of this lives on the M-CHAT-R/F page.
No authority prescribes a GARS-3 report format, but the instrument's nature as an informant rating dictates what a defensible results section must contain: who rated, under what conditions of contact, which index and why, the subscale pattern before the composite, the classification with its meaning attached, explicit convergence and divergence, and a bright line between diagnosis and eligibility. Each section below carries the pitfall that most often undermines it.
Identification: instrument, respondent, and index. Open with the full title and edition, the date completed, the language or version (English GARS-3 or Spanish GARS-3S), and the respondent by name or role, relationship, setting, and approximate duration and frequency of contact. Then name the composite actually generated, Autism Index 4 or Autism Index 6, and give the communication-status reason it was the applicable one. Pitfall: "Parent completed the GARS-3." Which parent, how much contact, which index? An informant score without its informant context is not interpretable, and "the Autism Index" without a 4 or 6 is ambiguous.
Rating conditions and informational validity. State whether the respondent had sufficient opportunity to observe the relevant behaviors, and record material limitations: a recent placement, contact limited to one highly structured setting, primarily virtual contact, interpreter use, or substantial unanswered items. This documents the conditions under which the informant formed the judgments, which is different from pronouncing the scores psychometrically valid. Pitfall: A bare "results are considered valid." The honest version describes the observation window and lets it bound the interpretation.
Subscale pattern before the composite. Describe which of the six domains carried the elevation, which were comparatively unremarkable, and whether the pattern was broad or concentrated, in original clinical language tied to concrete observations. The pattern is the clinically useful content; the composite only summarizes it. Pitfall: A results section that reads as an item inventory. Paraphrasing proprietary items drifts into reproduction; describe domains and behavior themes instead.
The index and classification, with meaning attached. Report the applicable index as a standard score (mean 100, SD 15) with the licensed classification from the scored protocol, then say immediately what it is: one respondent's standardized rating pattern resembling, or not resembling, the instrument's autism reference sample. The probability language is a category label, not a literal percentage chance. Pitfall: "The GARS-3 shows a 90 percent probability of autism." No such number exists; the classification is a band label, and presenting it as an individualized probability is the field's most repeated GARS-3 error.
Convergence and divergence, written out. State where the rating agrees and disagrees with direct observation, developmental history, interview, records, and any other informant, and what the disagreement means: context-dependence, setting demands, compensatory behavior, or a better competing explanation. A discrepancy is a clinical finding to investigate, not an error to average away. Pitfall: The GARS-3, observation, and interview each in their own silo with no sentence connecting them, or the higher score quietly chosen as the truth.
Diagnosis versus eligibility, kept separate. Write distinct conclusions: whether the full evidence supports DSM-5 autism (a clinical judgment integrating criteria, history, and observation) and whether the student meets the state's educational definition, which turns on adverse educational effect and need for specially designed instruction. The index satisfies neither by itself, and educational eligibility does not require a medical diagnosis in states like Texas, Minnesota, and Oregon. Pitfall: "The Autism Index supports eligibility." Eligibility is a team determination under the state rule; a score cannot make it, and a medical diagnosis is a different conclusion again.
Recommendations and boundaries. Close with recommendations that are individualized rather than imported: publisher-generated instructional objectives are suggestions requiring clinical and educational judgment, not IEP content to paste. State the boundaries: the scale contributes evidence, re-administration has no established reliable-change standard, and any change in scores describes changed informant report, to be read alongside functional data. Pitfall: "Severity improved by one category since last year." No reliable-change index or minimal important difference exists for the GARS-3; that sentence outruns the evidence.
GARS-3 RESULTS SECTION SKELETON Student/client: [initials] Age: [ ] Date completed: [ ] Version: [GARS-3 English / GARS-3S Spanish] Respondent: [name or role; relationship; setting; duration + frequency of contact] Index generated: [Autism Index 4 / Autism Index 6] because [communication status per the manual's rule] Rating conditions: [opportunity to observe; placement recency; settings known; interpreter; items unanswered] Subscale pattern (before the composite): Most prominent: [domains + concrete behavior themes] Comparatively unremarkable: [domains] Pattern: [broad / concentrated] Index result: [standard score, mean 100 SD 15] with the manual's classification: [licensed category from the scored protocol] Meaning: [one respondent's standardized rating pattern; a category label, not a percentage; not a diagnosis] Convergence / divergence: [agreement and disagreement with observation, history, interview, records, other informants; what the discrepancy means] Diagnostic conclusion (clinical): [supported / not supported / pending, from the full evidence] Educational eligibility: [team determination under the state definition; adverse effect + need documented separately] Recommendations: [individualized; publisher objectives adapted, never pasted; further data if divergent] Evaluator signature / credentials: Date:
Free to use and share, no signup. The PDF includes a one-page cheat sheet with section-by-section pitfalls and a pre-sign checklist; the DOCX is the blank results-section skeleton, ready to adapt. Neither reproduces test items, rating anchors, norms, or cut values.
Scenario: a school autism evaluation with a teacher-completed GARS-3 that converges with observation and history, written so the classification supports the team's work without becoming a diagnosis. All details are fictional.
Patient: D.K., 8 years · Setting: School multidisciplinary autism evaluation (initial) · Clinician: M. Reyes, PhD, NCSP, school psychologist · Note date: 08/20/2026
Identification and respondent context: Gilliam Autism Rating Scale, Third Edition (GARS-3), English version, completed 08/17/2026 by D.K.'s third-grade general-education teacher, who has taught him daily since the start of the school year and across structured instruction, small groups, and unstructured times. Because D.K. routinely uses functional phrase and sentence speech, all six subscales were applicable and Autism Index 6 was generated, per the manual's communication-status rule. The rating is one standardized informant measure within an evaluation that also includes structured observation, developmental history with his parents, a speech-language assessment, and record review.
Rating conditions: The teacher reported ample opportunity to observe the rated behaviors across the school day and completed all items; no interpreter was needed. Her contact is limited to the school setting, so her ratings describe D.K. at school, and the parent history below supplies the home perspective.
Subscale pattern and index: The teacher's ratings were most elevated on the Social Interaction, Social Communication, and Restrictive/Repetitive Behaviors subscales, with comparatively fewer concerns on Emotional Responses, Cognitive Style, and Maladaptive Speech: a concentrated pattern centered on reciprocal social behavior and sameness. The resulting Autism Index 6 fell within the manual's highest probability-of-autism classification. That classification means this teacher reported a pattern strongly resembling the behaviors represented in the GARS-3 autism reference sample; it is a category label from the scored protocol, not a percentage chance, and it does not by itself establish a diagnosis or eligibility. Exact values remain in the protocol; the manual's conversion tables are proprietary.
Convergence and divergence: The rating converged with direct observation across two settings, which documented limited spontaneous peer engagement, reduced conversational reciprocity, distress at an unannounced schedule change, and repetitive organization of materials, and with the developmental history, in which his parents described longstanding sameness needs and reduced reciprocal interaction in unfamiliar settings, while reporting somewhat fewer difficulties within familiar home routines. That home-school difference reads as contextual (predictable routines and one-to-one attention at home) rather than as disagreement about the pattern itself.
Conclusions and recommendations: Diagnostically, the converging evidence across informant rating, observation, history, and speech-language findings supports autism spectrum disorder as defined by DSM-5-TR; that conclusion rests on the full evidence, not on the GARS-3 score. Educationally, eligibility under the state's autism category is the team's determination and turns on adverse educational effect and need for specially designed instruction, documented in the eligibility section of this report. Recommendations were individualized from the evaluation as a whole; publisher-generated instructional objectives were reviewed as suggestions and adapted where they matched D.K.'s actual profile. Re-administration of the GARS-3 alone will not be used as a progress measure, because no reliable-change standard exists for it; progress will be monitored through IEP goal data and observation.
This sample is fictional and for educational purposes. It does not describe a real student or record; the details are invented to show write-up structure and are not clinical guidance. No items, rating anchors, norms, or cut values are reproduced.
Writing these after every session? BastionGPT drafts complete notes from bullets, dictation, or a transcript.
Generate a note from bulletsUnited States: the governing frames are specific and they all say the same thing about single measures. IDEA requires a variety of assessment tools, expressly prohibits any single measure or assessment as the sole criterion for eligibility or programming, and requires technically sound instruments used by trained personnel per producer instructions (34 CFR 300.304); the educational autism definition (300.8(c)(1)) turns on characteristics that adversely affect educational performance, which no rating scale can establish alone; and public agencies must consider independent educational evaluations that meet agency criteria (300.502), where consider does not mean adopt (LAW). States make the eligibility-versus-diagnosis line explicit: Texas rule 19 TAC 89.1040(c)(1) states that eligibility "cannot require that the student meets the requirements for a medical/psychological diagnosis of autism," Minnesota requires multiple methods, settings, observations, and developmental history and makes a medical diagnosis neither required nor sufficient, and Oregon's multidomain rule (amended December 2023) treats rating-scale information as one component (LAW, state by state). On the payer side, no reviewed policy accepts a GARS-3-only evaluation: TRICARE's autism program lists the GARS among validated instruments while requiring a diagnostic evaluation beyond a parent questionnaire (the program is authorized through December 2028), North Carolina Medicaid's policy amended August 1, 2026 names BOSA, TAP, ADOS-2, or CARS-2 for a nonprovisional qualifying diagnosis, with the GARS-3 absent, Aetna's autism bulletin lists the Gilliam among informant instruments inside a comprehensive evaluation, and Cigna's autism policy (effective May 2026, next review December 2026) recognizes comprehensive assessment rather than any index threshold; testing itself bills through the 96130 to 96139 family, and EPSDT obliges states to cover medically necessary child services without converting any score into an authorization (PAYER POLICY).
Canada and Australia put the same weight on multi-source assessment. The Canadian Paediatric Society's 2019 diagnostic guidance holds that no single tool makes the diagnosis and that standardized measures sit under clinical judgment; several jurisdictions add teeth, most concretely British Columbia, which requires both the ADI-R and the ADOS for publicly funded autism assessment, a pairing a GARS-type scale cannot substitute for (CONVENTION and provincial PAYER POLICY). Canada's earlier national best-practice guidelines went further and recommended against the original GARS on sensitivity grounds, part of why a Canadian report should present the GARS-3 strictly as supporting informant evidence with its respondent context intact. Australia's National Guideline for the Assessment and Diagnosis of Autism (2023 update, NHMRC-approved December 2023) prescribes comprehensive, needs-based assessment drawing on direct observation, information from people who know the person, developmental and medical history, and standardized tools where appropriate; a rating scale informs but never replaces that process, and report language that says so travels well across all three countries (CONVENTION). In every jurisdiction the portable artifact is the same: a write-up that preserves who rated, under what conditions, which index, the pattern, and how the evidence converged lets any downstream team, school, funding program, or diagnostician, reuse the evidence under its own rules.
Instrument facts, psychometric honesty, and rights complete a defensible page. Facts: the GARS-3 remains the current edition as of August 2026 (no GARS-4 exists or is announced); the manual is ©2014 though some distributor metadata says 2013; current publisher pages specify 56 items while several peer-reviewed papers say 58, a discrepancy to resolve against the licensed form rather than a web page; and the same product is qualification level B at PRO-ED and Pearson but level C at WPS, vendor purchasing classifications rather than properties of the test. Psychometrics: the manual reports sensitivity .97, specificity .97, and AUC .93 from a standardization sample of 1,859 collected 2010 to 2011, drawn entirely from individuals already diagnosed with autism; the strongest independent test, a 2022 study of 186 clinically referred children published in the Journal of Autism and Developmental Disorders, found the recommended cutoff correctly classified about 47 percent with false-positive rates of roughly 83 to 88 percent, a 2024 short-form follow-up could not optimize sensitivity and specificity at any cutoff, and the earlier GARS editions drew two decades of similar critique, so the manual's figures are manual findings, not settled field performance, and elevations in complex referral populations demand differential thinking (anxiety, ADHD, language disorder, trauma). No reliable-change standard exists, so re-administration is not a validated outcome measure. Rights: record forms are copyrighted consumables, reproduction and EHR embedding are separately licensed uses under PRO-ED's permissions process, the publisher's online system is the sanctioned scorer, and no authorized free public scorer exists; a report may always contain the examinee's derived scores and your interpretation. The GARS-3 and Gilliam Autism Rating Scale are product names of PRO-ED, Inc., and the instrument is copyrighted material of its publisher. BastionGPT is not affiliated with, or endorsed by, PRO-ED. This page reproduces no test items, rating anchors, norms, or cut values.
The numbers behind these errors are specific. The manual reports sensitivity and specificity of .97 from a 1,859-person standardization sample; an independent 2022 study of 186 clinically referred children found the recommended cutoff correctly classified about 47 percent, with false-positive rates near 83 to 88 percent; and North Carolina Medicaid's August 2026 policy names four qualifying diagnostic procedures, none of them the GARS-3. The BastionGPT Clinical Advisory Board sees the same errors most often in GARS-3 documentation reviews:
BastionGPT is specifically trained, tuned, and clinically tested on psychological and psychoeducational evaluation reports.
See how clinicians use it day to day on the AI therapy notes page.
Many BastionGPT users report saving more than 90 minutes per day on documentation.
HIPAA-compliant with a signed BAA on every plan. Your data is never used to train models. BastionGPT drafts, you review and sign.
Six subscale scaled scores (mean 10, SD 3), one per domain: Restrictive/Repetitive Behaviors, Social Interaction, Social Communication, Emotional Responses, Cognitive Style, and Maladaptive Speech. The applicable subscales then combine into a composite Autism Index on a mean-100, SD-15 metric: Autism Index 4 from the first four subscales when speech is absent or severely limited, or Autism Index 6 from all six when the examinee has sufficient functional speech. The manual maps the index to ordered probability-of-autism classifications and to a severity framing the publisher aligns with DSM-5 support levels; the exact bands and conversion tables are proprietary, so reports quote the licensed category from the scored protocol rather than reproducing tables. Two cautions travel with the numbers: the classification is a category label, not a percentage, and the severity framing is the instrument's interpretive label, not a clinical support-level determination, which is made clinically and can differ across DSM-5's two domains.
It means the respondent's ratings resemble, to a stated degree, the behavioral pattern in the instrument's norms, and nothing more precise than that. The norms were drawn entirely from individuals already diagnosed with autism, so the index expresses closeness to an autism reference sample as reported by one rater in one setting. A classification in the manual's highest band is not a statement that the person has a specific percentage chance of autism, and a lower band does not rule autism out (the original GARS line was criticized for missing nearly half of clinically diagnosed cases). A serviceable parent explanation: this is the category assigned to the pattern this respondent reported; it tells us how closely those ratings resemble the profile in the instrument's autism sample; it is not a percentage and it does not make the diagnosis by itself. The report then shows what converged, what diverged, and what additional evidence carried the conclusion.
The one the manual's communication-status rule selects, named explicitly with its rationale. Autism Index 4 uses the four subscales that do not depend on speech and applies when the examinee is nonverbal or has severely limited speech; Autism Index 6 adds Cognitive Style and Maladaptive Speech and applies when functional speech is sufficient for all six domains to be rated meaningfully. They are not interchangeable options to report side by side, and they are not parent and teacher versions: when multiple informants rate the same examinee, every rating uses the index matching that examinee's communication status. The write-up error to avoid is the unlabeled "Autism Index," which leaves the reader unable to know which composite, and which subscales, produced the number; the indefensible error is switching indexes because one produces a preferred result.
No, by the publisher's own framing and by the independent evidence. PRO-ED describes the scale as assisting in identifying autism and estimating severity, and the manual builds in a diagnostic validation step mapping the presentation to DSM-5 criteria, both signals that the rating supports rather than makes the determination. The independent numbers close the question: while the manual reports .97 sensitivity and specificity, a 2022 study of 186 clinically referred children found the recommended cutoff correctly classified about 47 percent, with false-positive rates around 83 to 88 percent, because the behaviors it rates also occur in language disorder, intellectual disability, ADHD, anxiety, and trauma-related presentations. A diagnosis integrates direct observation (the ADOS-2), developmental history and interview (the ADI-R), clinical judgment against DSM-5-TR criteria, and differential diagnosis; the GARS-3 contributes one informant's standardized evidence to that synthesis.
No. Federal regulations require a variety of assessment tools and prohibit any single measure or assessment from serving as the sole criterion for determining eligibility, and the educational autism definition turns on characteristics that adversely affect educational performance, something no rating scale can establish. States say the rest plainly: Texas rule 19 TAC 89.1040(c)(1) states that eligibility "cannot require that the student meets the requirements for a medical/psychological diagnosis of autism," Minnesota requires multiple methods across settings plus systematic observation and developmental history, with a medical diagnosis neither required nor sufficient, and Oregon's rule makes rating-scale information one component of a multidomain evaluation. So the GARS-3 can help satisfy part of one method; the eligibility determination belongs to the team applying the state definition, documenting educational effect and the need for specially designed instruction. The same logic runs in reverse: a modest index does not defeat eligibility that the fuller evidence supports.
As a component, sometimes; alone, no policy reviewed for this page does. TRICARE's autism program lists the GARS among validated instruments a diagnosing provider may use, while stating that a parent questionnaire alone is insufficient and requiring a diagnostic evaluation from a qualified provider. North Carolina Medicaid's policy amended August 1, 2026 goes further for a nonprovisional diagnosis, naming BOSA, TAP, ADOS-2, or CARS-2 as qualifying diagnostic procedures, a list that does not include the GARS-3, and rejecting decisions based solely on screening tools or educational determinations. Aetna's autism bulletin lists the Gilliam among informant instruments within a comprehensive evaluation, and Cigna's policy recognizes comprehensive assessment rather than any index threshold. The practical rule for the write-up: present the GARS-3 as standardized informant evidence inside a comprehensive diagnostic evaluation, name the rest of the battery, and check the member's current policy rather than asserting the scale is insurance-accepted.
State the divergence directly, then reason through it instead of averaging it away. Neither result automatically invalidates the other: a brief structured observation cannot rule autism out, and an elevated informant scale cannot establish it. Work the plausible explanations in writing: behavior may be context-dependent (the structured session reduced the demands that trigger difficulties), the respondent may observe situations testing never reproduces, compensatory behavior or masking may operate in session, or the elevation may reflect difficulties better explained by anxiety, ADHD, language disorder, or trauma, which the false-positive literature makes a live possibility in referred populations. Then say what decides it: developmental history, additional settings and informants, language and cognitive findings, adaptive functioning, and differential diagnosis. The conclusion may be that autism is supported despite the observation, unsupported despite the rating, or unresolved pending more data; what makes the report defensible is showing the reconciliation rather than picking the score that matches a preferred answer.
Not on its own. No independent reliable-change index, minimal clinically important difference, or treatment-response benchmark has been established for the GARS-3, so there is no defensible way to say a given score difference exceeds measurement noise, and "severity improved by one category" claims more than the instrument can support. A repeated administration documents change in one informant's report, which is worth describing: the same respondent, the same index, and the interval should all be stated, and the change should sit beside the evidence that can carry a progress claim, direct observation, IEP goal and service data, adaptive-behavior measures, and functional outcomes. If the respondent changed between administrations, say so prominently, because informant effects alone can move rating-scale scores. For payers and teams, frame re-administration as updated informant perception within a broader progress review, never as the outcome measure.
Yes. Give it the facts (respondent with relationship and contact time, the index used and its communication-status rationale, the subscale pattern, the licensed classification, observation and history findings, and the eligibility or diagnostic framework in play) and it drafts the results section: informant context first, pattern before composite, the classification defined rather than inflated, convergence and divergence written out, and the diagnostic and eligibility conclusions kept separate, ready for your review. It can also cross-check a finished report for an unnamed index, missing respondent context, percentage language, manual figures quoted as field performance, a score doing eligibility work, or an unsupported progress claim, and it can draft the parent explanation and divergence-reconciliation paragraphs. BastionGPT is HIPAA-compliant with a signed BAA on every plan, and your data is never used to train models.
The instrument facts and compliance claims on this page trace to these sources, last verified August 2026:
Educational content, not legal or billing advice. Sample notes are fictional. Follow your organization's policies and your board, payer, and jurisdiction requirements.