A cluster is a mini-epidemic of a rare disease. Clusters may give clues to the etiology of disease, or may signal a hazardous exposure. Unfortunately, cluster investigations seldom are conclusive, for several reasons. Statistically significant clusters can occur by chance. The probability of finding chance cancer clusters is calculated for the 200 Alaska Native villages. The problem of selection bias is explained, and other limitations of epidemiology are described. A logical, stepwise protocol for investigating cancer clusters is presented.