BACKGROUND: It is not known whether early intervention can improve long-term autism symptom outcomes. We aimed to follow-up the Preschool Autism Communication Trial (PACT), to investigate whether the PACT intervention had a long-term effect on autism symptoms and continued effects on parent and child social interaction. METHODS: PACT was a randomised controlled trial of a parent-mediated social communication intervention for children aged 2-4 years with core autism. Follow-up ascertainment was done at three specialised clinical services centres in the UK (London, Manchester, and Newcastle) at a median of 5·75 years (IQR 5·42-5·92) from the original trial endpoint. The main blinded outcomes were the comparative severity score (CSS) from the Autism Diagnostic Observation Schedule (ADOS), the Dyadic Communication Assessment Measure (DCMA) of the proportion of child initiatiations when interacting with the parent, and an expressive-receptive language composite. All analyses followed the intention-to-treat principle. PACT is registered with the ISRCTN registry, number ISRCTN58133827. FINDINGS: 121 (80%) of the 152 trial participants (59 [77%] of 77 assigned to PACT intervention vs 62 [83%] of 75 assigned to treatment as usual) were traced and consented to be assessed between July, 2013, and September, 2014. Mean age at follow-up was 10·5 years (SD 0·8). Group difference in favour of the PACT intervention based on ADOS CSS of log-odds effect size (ES) was 0·64 (95% CI 0·07 to 1·20) at treatment endpoint and ES 0·70 (95% CI -0·05 to 1·47) at follow-up, giving an overall reduction in symptom severity over the course of the whole trial and follow-up period (ES 0·55, 95% CI 0·14 to 0·91, p=0·004). Group difference in DCMA child initiations at follow-up showed a Cohen's d ES of 0·29 (95% CI -0.02 to 0.57) and was significant over the course of the study (ES 0·33, 95% CI 0·11 to 0·57, p=0·004). There were no group differences in the language composite at follow-up (ES 0·15, 95% CI -0·23 to 0·53). INTERPRETATION: The results are the first to show long-term symptom reduction after a randomised controlled trial of early intervention in autism spectrum disorder. They support the clinical value of the PACT intervention and have implications for developmental theory. FUNDING: Medical Research Council.