Background The UK guideline recommends 3-yearly surveillance for patients with intermediate-risk (IR) adenomas. No study has examined whether or not this group has heterogeneity in surveillance needs. Objectives To examine the effect of surveillance on colorectal cancer (CRC) incidence; assess heterogeneity in risk; and identify the optimum frequency of surveillance, the psychological impact of surveillance, and the cost-effectiveness of alternative follow-up strategies. Design Retrospective multicentre cohort study. Setting Routine endoscopy and pathology data from 17 UK hospitals (n = 11,944), and a screening data set comprising three pooled cohorts (n = 2352), followed up using cancer registries. Subjects Patients with IR adenoma(s) (three or four small adenomas or one or two large adenomas). Primary outcomes Advanced adenoma (AA) and CRC detected at follow-up visits, and CRC incidence after baseline and first follow-up. Methods The effects of surveillance on long-term CRC incidence and of interval length on findings at follow-up were examined using proportional hazards and logistic regression, adjusting for patient, procedural and polyp characteristics. Lower-intermediate-risk (LIR) subgroups and higher-intermediate-risk (HIR) subgroups were defined, based on predictors of CRC risk. A model-based cost–utility analysis compared 13 surveillance strategies. Between-group analyses of variance were used to test for differences in bowel cancer worry between screening outcome groups (n = 35,700). A limitation of using routine hospital data is the potential for missed examinations and underestimation of the effect of interval and surveillance. Results In the hospital data set, 168 CRCs occurred during 81,442 person-years (pys) of follow-up [206 per 100,000 pys, 95% confidence interval (CI) 177 to 240 pys]. One surveillance significantly lowered CRC incidence, both overall [hazard ratio (HR) 0.51, 95% CI 0.34 to 0.77] and in the HIR subgroup (n = 9265; HR 0.50, 95% CI 0.34 to 0.76). In the LIR subgroup (n = 2679) the benefit of surveillance was less clear (HR 0.62, 95% CI 0.16 to 2.43). Additional surveillance lowered CRC risk in the HIR subgroup by a further 15% (HR 0.36, 95% CI 0.20 to 0.62). The odds of detecting AA and CRC at first follow-up (FUV1) increased by 18% [odds ratio (OR) 1.18, 95% CI 1.12 to 1.24] and 32% (OR 1.32, 95% CI 1.20 to 1.46) per year increase in interval, respectively, and the odds of advanced neoplasia at second follow-up increased...