Epidemiology traditionally has relied on measures of "external" exposure in determining the association between exposure and disease. Recently, there has been increasing reliance on internal markers reflecting internal dose and/or early stages of disease. In the context of observational studies of chronic disease in which there is a known exposure-disease association, the question arises whether the external exposure or the internal marker is a better predictor of eventual disease outcome. Here we describe some simple approaches to evaluate the relative predictive value of the internal marker (or biomarker, defined in the most general sense) versus the exposure, as well as their limitations. The problems of assessing the predictive value of internal markers for chronic disease are illustrated via two examples: (a) carcinogens, cytogenetic outcomes, and cancer; and (b) asbestos, asbestosis, and lung cancer. We conclude that it is unlikely that observational epidemiology will allow a full assessment of the predictive value of cytogenetic outcomes versus exposure for cancer in humans exposed to known carcinogens in the near future, although animal studies could provide important complementary information. For asbestos, data to date indicate that the presence or absence of asbestosis is a better predictor of lung cancer in an exposed population than is the level of exposure to asbestos itself. In general, the most useful markers for predicting chronic disease are ones which persist over time.