This paper discusses some optimum design problems which arise when paging and segmentation are used in a multiprogrammed computer system, and suggests approaches to their solution. It is argued that design optimization depends fundamentally on the statistics of the typical software load presented to the computer system, and also that the investigation requires the aid of a total system model. Statistical techniques for defining software characteristics in such a model are presented. It is shown how these tools can be employed to simulate the changing contents of main memory; and as a further illustration, the simulation of an auxiliary memory is demonstrated. Graphs are given which show the variation of auxiliary memory performance with the number of associative cells available.