BACKGROUND: Human immunodeficiency virus (HIV) prevalence in Uganda fell during the 1990s and has risen since 2000. The changing trends since 2000 may be due to changing risky sexual behaviour, but other factors may also contribute. We explore the possible impact that two factors may have on cohort trends: natural epidemic dynamics and local migration. We simultaneously fit a mathematical model to the contrasting prevalence trends by age/gender in a southwest Ugandan cohort, which has never been done before. METHODS: We fit a model to HIV trends, by age groups and gender, assuming: (i) neither migration nor intentional behaviour change, (ii) migration changes, (iii) intentional behaviour changes and (iv) both change. The model fits were assessed through sum of squares goodness of fits. We also explored the impact of the natural course of the epidemic on average partner acquisition rates, under the condition of no intentional behaviour change over time. RESULTS: Without allowing intentional change in partner acquisition rates, an excellent fit was achievable to prevalence trends by gender. However, all of the contrasting trends by age/gender could not be replicated simultaneously. Adding intentional behaviour change improved the fit to some groups, but not all. Adding both intentional behaviour change and migration further improved the fit. CONCLUSIONS: While some of the increasing HIV prevalence in Uganda since 2000 may be due to increased risky behaviour, some of the observed epidemiologic trends would likely occur without any intentional change in behaviour. Average population-level behaviour can change due to preferential mortality among higher risk takers, without individuals changing their behaviour.