This paper reviews a number of past studies in the field of diffusion research, describing the major features of each diffusion model and discussing its value for predicting the spread of educational innovations. Following this review, the author presents a new autocatalytic diffusion model based on the mathematical models of epidemiologists and chemists. This autocatalytic model is adapted to the study of educational innovations, and then the model is applied to historical data and used to predict the life cycle of six different educational innovations. The predicted life cycles matched the actual historical data very well in five of the six cases. The author also uses the model to project the future adoption life cycle of the semester system in Ontario secondary schools, an innovation that is still in the process of adoption. (JG)