There is little evidence that neoadjuvant androgen deprivation therapy (ADT) of 3 months' duration before radical prostatectomy (RP) favorably influences disease-free survival. However, recent data suggest that prolonged treatment may improve outcome. We conducted a prospective cohort study to determine whether ADT of either standard or prolonged duration before RP influences the risk of prostate-specific antigen (PSA) failure. We followed 756 men treated for prostate cancer by RP between 1991 and 1998 in Quebec City. Of these, 240 received combined neoadjuvant ADT for either </=92 days (129 men) or >/=93 days (111 men), and 516 were treated by RP alone. Multivariate Cox regression was used to estimate the hazard ratios (HR) of PSA failure (>0.3 ng/mL) associated with treatment regimen controlling for age, clinical stage, grade, and initial PSA level. The median duration of follow-up was 4 years. Compared with men treated by RP alone, those who received neoadjuvant ADT for >/=93 days had an HR of PSA failure of 0.60. The inverse association with the risk of PSA failure became statistically significant from the third year on, reached its greatest magnitude after 4 years, and was still present 8 years after RP. No association was observed for ADT of </=92 days. These results suggest that neoadjuvant ADT before RP has a real, delayed, and persistent effect on disease-free survival, if and only if ADT is prolonged beyond 3 months.