BACKGROUND: Human papillomavirus (HPV) vaccines were first licensed as a three-dose series. Two doses are now widely recommended in some age groups; there are data suggesting high efficacy with one dose. We updated a systematic literature review of HPV vaccine effectiveness by number of doses in observational studies. METHODS: We searched Medline and Embase databases from January 1, 2007, through September 29, 2021. Data were extracted and summarized in a narrative synthesis. We also conducted quality assessments for bias due to selection, information, and confounding. RESULTS: Overall, 35 studies were included; all except one were conducted within the context of a recommended three-dose schedule. Evaluations were in countries that used bivalent HPV vaccine (seven), quadrivalent HPV vaccine (27) or both (one). Nine evaluated effectiveness against HPV infection, ten anogenital warts, and 16 cervical abnormalities. All studies were judged to have moderate or serious risk of bias. The biases rated as serious would likely result in lower effectiveness with fewer doses. Investigators attempted to control for or stratify by potentially important variables, such as age at vaccination. Eight studies evaluated impact of buffer periods (lag time) for case counting and 10 evaluated different intervals between doses for two-dose vaccine recipients. Studies that stratified by vaccination age found higher effectiveness with younger age at vaccination, although differences were not all formally tested. Most studies found highest estimates of effectiveness with three doses; significant effectiveness was found among 28/29 studies that evaluated three doses, 19/29 that evaluated two doses, and 18/30 that evaluated one dose. Some studies that adjusted or stratified analyses by age at vaccination found similar effectiveness with three, two and one doses. CONCLUSION: Observational studies of HPV vaccine effectiveness have many biases. Studies examining persons vaccinated prior to sexual activity and using methods to reduce sources of bias are needed for valid effectiveness estimates.