BACKGROUND: Human papillomavirus (HPV) testing on self-collected samples is a potential alternative to HPV testing on clinician-collected samples, but non-inferiority of its clinical accuracy remains to be assessed in the regular screening population. The IMPROVE study was done to evaluate the clinical accuracy of primary HPV testing on self-collected samples within an organised screening setting. METHODS: In this randomised, non-inferiority trial, women aged 29-61 years were invited to participate in the study as part of their regular screening invitation in the Netherlands. Women who provided informed consent were randomly allocated (1:1, with a block size of ten stratified by age) to one of two groups: a self-sampling group, in which women were requested to collect their own cervicovaginal sample using an Evalyn Brush (Rovers Medical Devices BV, Oss, Netherlands); or a clinician-based sampling group, in which samples were collected by a general practitioner with a Cervex-Brush (Rovers Medical Devices BV). All samples were tested for HPV using the clinically validated GP5+/6+ PCR enzyme immunoassay (Labo Biomedical Products BV, Rijswijk, Netherlands). HPV-positive women in both groups were retested with the other collection method and triaged by cytology and repeat cytology in accordance with current Dutch screening guidelines. Primary endpoints were detection of cervical intraepithelial neoplasia (CIN) of grade 2 or worse (CIN2+) and grade 3 or worse (CIN3+). Non-inferiority of HPV testing on self-collected versus clinician-collected samples was evaluated against a margin of 90% for the relative sensitivity and 98% for the relative specificity. This study is registered at the Dutch Trial register (NTR5078) and has been completed. FINDINGS: Of the 187 473 women invited to participate, 8212 were randomly allocated to the self-sampling group and 8198 to the clinician-based sampling group. After exclusion of women who met the exclusion criteria or who did not return their sample, 7643 women were included in the self-sampling group and 6282 in the clinician-based sampling group. 569 (7·4%) self-collected samples and 451 (7·2%) clinician-collected samples tested positive for HPV (relative risk 1·04 [95% CI 0·92-1·17]). Median follow-up duration for HPV-positive women was 20 months (IQR 17-22). The CIN2+ sensitivity and specificity of HPV testing did not differ between self-sampling and clinician-based sampling (relative sensitivity 0·96 [0·90-1·03]; relative specificity 1·00 [0·99-1·01]). For the CIN3+ endpoint, relative sensitivity was 0·99 (0·91-1·08) and relative specificity was 1·00 (0·99-1·01). INTERPRETATION: HPV testing done with a clinically validated PCR-based assay had similar accuracy on self-collected and clinician-collected samples in terms of the detection of CIN2+ or CIN3+ lesions. These findings suggest that HPV self-sampling could be used as a primary screening method in routine screening. FUNDING: Ministry of Health, Welfare, and Sport (Netherlands), and the European Commission.