Using a multiplicative model, the age-specific relative risk of infection with the human immunodeficiency virus, HIV, and the time-dependent HIV infection rate can be estimated simultaneously from acquired immunodeficiency syndrome, AIDS, incidence data. Non-parametric estimates are arrived at by an EM-algorithm, with smoothing, applied to a back-projection model. The method gives estimates which lie on a smooth curve in accordance with strong prior expectations. The approach is illustrated with an application to Australian AIDS data. It is found that the HIV infection rate tends to be estimated more precisely when age is included as a covariate.