Abstract Background Studies comparing self-collected vaginal samples with clinician-collected cervical samples with respect to high-risk human papillomavirus (HPV) detection and genotype agreement based on clinically validated full HPV genotype assays (e.g. the CLART HPV4S) are limited. This study compared the two types of samples using the CLART assay with respect to HPV detection and genotype agreement in a referral population. Methods A total of 212 women aged 30–59 years and diagnosed with atypical squamous cells of undetermined significance (ASC-US) within the Danish cervical cancer screening programme had a cervical sample taken at their general practitioner. Afterwards, the women took a vaginal sample with the Evalyn Brush device at home. The paired samples were HPV-tested with the full genotyping CLART HPV4S assay. Histological results, i.e. cervical intraepithelial neoplasia of grade 2 or worse (CIN2+) were available for 14 women with HPV-positive clinician-collected samples. Results The study found the same HPV prevalence in self-collected vaginal samples compared to clinician-collected cervical samples (19.3%, 95% CI 14.3–25.3% vs 18.4%, 95% CI 13.4–24.3%). The CLART HPV4S assay detected approximately the same number of CIN2+ cases in the self-collected vaginal samples compared to the clinician-collected cervical samples (13 vs 11 cases). Exactly the same genotypes were detected in 75% (21/28) of the HPV-positive paired samples, while at least one identical genotype was found in the remaining 25% (7/28) of the paired samples. Conclusions The CLART HPV4S assay performed similarly well in self-collected vaginal samples as in clinician-collected cervical samples with respect to both HPV detection and genotype agreement when using the Evalyn Brush and the CLART HPV4S assay in a referral population. Although further evaluation is needed, the findings suggest that full HPV genotyping based on the CLART assay can be useful when establishing HPV genotype-specific referral strategies for women tested HPV-positive by self-sampling.