Objective: To prospectively identify psychosocial predictors of pregnancy among African-American adolescent females. Methods: African-American females, 14-18 yrs old, were recruited from schools and health clinics in low-income neighborhoods. Adolescents completed an in-depth survey and provided urine specimens for pregnancy testing at baseline and 6-month intervals for 1.5 years. Selected problem behaviors, demographic, and psychosocial variables were tested for bivariate and multivariate significance relative to biologically confirmed pregnancy during the follow-up period. Only adolescents who initially tested negative for pregnancy were included (n = 241). Results: About 26% (n = 63) of the adolescents became pregnant over the follow-up period. Although a broad spectrum of variables achieved bivariate significance, few retained significance in the multivariate model. Multivariate predictors of pregnancy were biologically confirmed marijuana use (adjusted odds ratios [AOR] = 12.4, P = 0.0003) and perceiving that the sex partner desired pregnancy (AOR = 1.27, P = 0.01). A protective effect was observed for adolescents who reported that a family member received welfare benefits; these adolescents were about 60% less likely to become pregnant (AOR = 0.38, P = 0.04). Conclusions: Pediatricians and other health professionals who participate in community efforts to prevent first and subsequent adolescent pregnancies may benefit from recognizing that marijuana use and pregnancy may be co-occurring problems. Adolescents' perceptions of their boyfriends' level of desire for conception may also be an important predictor of pregnancy risk. The findings also suggest a possible protective effect of receiving TANF (Temporary Assistance to Needy Families) benefits; adolescent recipients of these programs may be more vigilant in their pregnancy prevention practices than those who are not recipients.