The evolution of an HIV epidemic involves many population dynamics that are closely linked with age. Understanding these dynamics can help policy makers and program planners design interventions, measure progress in the fight against HIV, and prepare for the future impact of the disease. To this end, we propose a model for sex- and age-specific projections of HIV prevalence that uses B-splines to model changes in HIV incidence. We are able to combine HIV prevalence data from antenatal clinics with prevalence data from nationally representative surveys to estimate the model parameters. Several approaches are also proposed for generating forecasts from our relatively parsimonious B-spline model. We apply these techniques to investigate the dynamics of HIV incidence in Tanzania. The results support a rapid increase in HIV incidence from the early 1980s to the early 1990s, followed by a decline up through 2003. We also generate forecasts of HIV prevalence for 2007 and 2011, and compare them to nationally representative data from these years. Our assessment of the predictive performance of the approaches suggests that additional information is needed to produce accurate forecasts of sex- and age-specific prevalence, but that B-splines provide a potentially useful approach for achieving this goal.