The aim of this paper was to evaluate the factors that predict regression of untreated CIN 2 and 3. A total of 93 patients with colposcopic persistent CIN 2 and 3 lesions after biopsy were followed for 6 months. Human papillomavirus (HPV) types were determined by polymerase chain reaction at enrolment. We analysed the biologic and demographic predictors of natural regression using univariate and multivariate methods. The overall regression rate was 52% (48 out of 93), including 58% (22 out of 38) of CIN 2 and 47% (26 out of 55) of CIN 3 lesions (P=0.31 for difference). Human papillomavirus was detected in 84% (78 out of 93) of patients. In univariate analysis, 80% (12 out of 15) of lesions without HPV regressed compared to 46% (36 out of 78) of lesions with HPV infection (P=0.016). Women without HPV and those who had a resolution of HPV had a four-fold higher chance of regression than those with persistent HPV (relative odds=3.5, 95% CI=1.4-8.6). Women with five or fewer lifetime sexual partners had higher rates of regression than women with more than five partners (P=0.003). In multivariate analysis, HPV status and number of sexual partners remained as significant independent predictors of regression. In conclusion, HPV status and number of lifetime sexual partners were strongly predictive of regression of untreated CIN 2 and 3.