The concept of latency period has developed in chronic disease epidemiology by analogy with the incubation period associated with infectious diseases. We discuss the use of the concept of mean latency period to describe the temporal development of cancer and the underlying conceptual and methodologic problems. The respective roles of age at diagnosis, age at start of exposure, duration of exposure, and years since exposure in modifying cancer risks are considered. Multistep models for the development of cancer are illustrated, and their use and limitations for interpreting temporal associations in cancer development are discussed. Rather than trying to define exposure-disease relationships as being characterized by a long or a short latency period, it seems more promising to study the temporal sequence of disease development as a multistep process.