Autism is difficult to identify in adults due to lack of validated self-report questionnaires. We compared the effectiveness of the Autism Quotient (AQ) and the Ritvo Autism-Asperger's Diagnostic Scale-Revised (RAADS-R) questionnaires in adult mental health services in two English counties.A subsample of adults who completed the AQ and RAADS-R were invited to take part in an Autism Diagnostic Observation Schedule (ADOS Module 4) assessment with probability of selection weighted by scores on the questionnaires.364 men and 374 women consented to take part. Recorded diagnoses were most commonly mood disorders (44%) and mental and behavioural disorders due to alcohol/substance misuse (19%). 4.8% (95% CI 2.9-7.5) were identified with autism (ADOS Module 4 10+). One had a pre-existing diagnosis of autism; 26% (5) had borderline personality disorders (all female) and 17% (3) mood disorders. The AQ and RAADS-R had fair test accuracy (area under Receiver Operating Characteristics (ROC) curve 0.77 and 0.79 respectively). AQ sensitivity was 0.79 (95% CI 0.54-0.94) and specificity 0.77 (95% CI 0.65-0.86); RAADS-R sensitivity was 0.75 (95% CI 0.48-0.93) and specificity 0.71 (95% CI 0.60-0.81).The AQ and RAADS-R can guide decisions to refer adults in mental health services to autism diagnostic services.