Three methods for projecting the short-term course of the acquired immunodeficiency syndrome (AIDS) epidemic are discussed: (a) empirical extrapolation, (b) the method of "back calculation," and (c) projections based on compartmental models. Extrapolation, which requires only data on AIDS incidence, is based on an assumed functional form and on the supposition that previous trends will continue. The method of back calculation incorporates both information on previous AIDS incidence and knowledge about the incubation period distribution. These calculations provide some evidence of how many infections occurred during previous time intervals. Although this information is not precise, particularly for the recent past, it is sufficient to produce stable short-term projections. Compartmental models can be used to project future prevalence of infection as well as future AIDS incidence. However, such projections are very dependent on assumptions about initial numbers of individuals infected, rates of transmission, changes in high-risk behaviors over time, and assumptions about transmission among subpopulations with differing transmission rates and initial prevalence of infection. Thus, compartmental models offer insights into the trends in an epidemic but do not currently provide a practical tool for obtaining quantitative projections. We present projections for various risk groups based on the method of back calculation and discuss the use of additional epidemiologic data to obtain accurate projections a decade in advance.