Background The aims of this dissertation were to study intimate partner violence among adolescents in Tanzania and South Africa, particularly prevalences and associated factors. Furthermore, to examine the relation between violent attitudes and violent behaviour, and the importance of the Theory of Planned Behaviour in predicting sexual debut. Intimate partner violence is the most common form of violence against women and girls. About one third of all women worldwide will experience violence from an intimate partner at least once in their lifetime. This has many negative effects on the physical, mental and reproductive health of women, among which a heightened risk for infection with HIV. This is especially worrying given the high prevalence of HIV in sub-Saharan Africa. Infection can be direct, through sexual violence and rape. Infection can also be indirect: for example, a culture of masculinity that condones male control of women, male sexual entitlement, patriarchy, and perceived low female status, all contribute to ‘justifying’ violence against women, thereby increasing chances of HIV infection. With a few exceptions, the majority of studies from Africa on violence against intimate partners, concerns people from college age and onwards, and are cross-sectional in nature. Various social cognition models like the Theory of Planned Behaviour have been used in such research. On the one hand these models have shown to be useful in various Africa settings; on the other hand there is also resistance against use of such Western models in the ‘global south’. The current prospective, longitudinal study will contribute to expanding knowledge on intimate partner violence among adolescents (12-15 year olds), including the usefulness of social cognition models, specifically in Tanzania and South Africa. Methods The three papers in this study were based on data material from survey data collections done in the SATZ study research project, a large-scale randomised controlled trial done in Tanzania and South Africa(n=15,864) between 2002-2006. Analyses in the first two papers were carried out with SPSS version 15.0, and included descriptive statistics, cross-tabulations, factor analyses, logistic regressions and general linear modelling. Analyses in paper III were done in SPSS version 19.0 and Mplus version 6, and included multi-group structural equation modelling, in addition to descriptive statics. All analyses were adjusted for cluster effects. Measurement included scales on intimate partner violence (Straus’ Revised Conflict Tactics Scales, CTS2), sexual debut, attitude towards violence, social cognition scales, and sociodemographic variables. Results Paper I. Violence was prevalent in all sites: depending on site and gender, 10.2-37.8% has been victim, 3.1-21.8% perpetrators, and 8.6-42.8% has been both victim and perpetrator. Being male was associated with perpetration in all sites, while being female was associated with victimization in Dar es Salaam and Mankweng (yet in Cape Town victimization was also associated with being male). In all sites higher age and low socioeconomic status was associated with all types of violence. In Cape Town, being religious and having parents with higher education was protective against all types of violence. However no clear subgroups were defined as being at much higher risk, so violence control policies should target young adolescents across geographic, economic and social groups. Paper II. We found that results were consistent with the notion of a bi-directional attitudes-behaviour interrelationship in Cape Town (and to some extent Mankweng). In Dar es Salaam attitudes predicted behaviour prospectively, but prediction in the opposite direction was not confirmed. Paper III. The results showed that the Theory of Planned Behaviour predictors did not have strong predictive power. Controlling for sociodemographic factors did not change the predictive power much, nor when violence was added to the model. However, violence did predict sexual debut directly, and explained variance in sexual debut was substantially higher when violence was added to the model. Discussion This study shows that intimate partner violence is highly prevalent, already at a young age, in relationships between young people in South Africa and Tanzania. To some extent this study found support for including individual attitude-and behaviour change approaches in intervention efforts: in paper II we found that in Cape Town (and to some extent Mankweng) there was a bidirectional relationship between attitudes and behaviour (attitudes towards violence and being a perpetration), while in Dar es Salaam attitudes predicted behaviour prospectively. In paper III support was found for prediction of intentions by attitudes (and to some extent behaviour by intentions). Still, the social cognition model could only explain a limited amount of explained variance; therefore there is a need to look into other environmental influences. We emphasize the need to focus on violence in the wider community at the larger social and structural level, beyond the individual level. Future research and interventions should include both sexes, since both are part of the dynamic that keeps violence going, and target young people, preferably before they become involved in (sexual) relationships. Interventions should be adjusted to the cultural and social settings where they take place, using an ecological approach where the individual is targeted, as well as their surroundings at the interpersonal, community, and structural level.