Although hardly new, caution is warranted in the interpretation of new findings in epidemiology as they are often due to chance. Replication remains the most important way to evaluate such findings. Although the application of statistical probability will be particularly useful in the identification of false-positive findings in genetic epidemiology, the problems of confounding and measurement error are uniquely small in this field. In other applications, epidemiology will be most successful when it is closely integrated with other areas of human biology.