Both experimental and observational studies of cancer need to have an end point. Traditionally, in aetiological and prevention studies, that end point has been the incidence of cancer itself, whereas in therapeutic trials, the end point is usually time to cancer recurrence or death. But cancer takes a long time to develop in an individual and is rare in the population. Therefore, aetiological studies and prevention trials must be large and lengthy to be meaningful. Similarly, many therapeutic trials require a long follow-up of large numbers of patients. Surrogate end points--markers of preclinical cancer or of imminent recurrence--are therefore an attractive alternative. But how can we be sure that a study with a surrogate outcome gives us the right answer about the true end point?