Population projections are simply extrapolations of demographic patterns that have remained constant in the past into the future. This observation by Keyfitz simultaneously provides a philosophical base for forecasting techniques and sets off a search for demographic patterns. We begin with a discussion of the reasons for disaggregate projections, how the reasons effect data requirements, and how models relieve the strain. The next section discusses advances in demographic models, especially extensions of the relational methods developed by Brass. Finally, we discuss how time-series models, in conjunction with model patterns, can be used to make forecasts with appropriate confidence intervals. The paper motivates the techniques through examples of Swedish life tables, and describes appropriate mathematical properties for projection models.