This study is a process evaluation of the school component of the adolescent sexual health programme MEMA kwa Vijana (MkV), which was implemented in 62 primary schools in rural Mwanza, Tanzania from 1999 to 2001. The MkV curriculum was a teacher-led and peer-assisted programme based on the Social Learning Theory. Process evaluation included observation of training sessions, monitoring and supervision, annual surveys of implementers, group discussions and 158 person-weeks of participant observation. Most teachers taught curriculum content well, but sometimes had difficulty adopting new teaching styles. Peer educators performed scripted dramas well, but were limited as informal educators and behavioural models. The intervention appeared successful in addressing some cognitions, e.g. knowledge of risks and benefits of behaviours, but not others, e.g. perceived susceptibility to risk. MkV shared the characteristics of other African school-based programmes found to be successful, and similarly found significant improvements in self-reported behaviour in surveys. However, a substantial proportion of MkV survey self-reports were inconsistent, there was no consistent impact on biological markers and extensive process evaluation found little impact on several key theoretical determinants of behaviour. Improvements in self-reported survey data alone may provide only a very limited-and perhaps invalid-indication of adolescent sexual health programme success.