• It has long been known that if the probability of dying of a given cancer remains constant over time, the logarithmic plot of survival of a group of patients with that cancer will be a straight line. More frequently, however, the probability of dying varies with time and the logarithmic survival plot is a curve. We believe that in some cases a population of patients with one type of neoplasm may be composed of several subgroups, each of which has a simple predictable course with a rectilinear survival plot. This article describes a method for partitioning such survival plots and identifying and characterizing the subgroups, and applies the method to a group of patients with brain tumors. The value of being able to identify these subgroups and their relative proportions, the observations that can be made with this analysis regarding the nature of the various subgroups of brain tumors, and the implications of changes in the half-lives as well as the proportions of the subgroups with time are discussed.