In the Chicago Western Electric Company study, diet was assessed at the initial examination, in 1957-1958, of 1900 middle-aged men and again at their second examination about one year later. At the first examination, lipid composition of the diet, as summarized by a score based on the formula of Keys, Anderson and Grande (Grande, F. Predicting change in serum cholesterol from change in lipid composition of the diet. In: Lauer RM, Shekelle RB, eds. Childhood Prevention of Atherosclerosis and Hypertension. New York: Raven Press, 1980:145-53), was positively associated with level of serum cholesterol. Between the first and second examinations, however, hypercholesterolemic men were more likely than others to have reduced intake of dietary saturated fatty acids and cholesterol. As a result, at the second examination the cross-sectional linear association between the diet score and serum cholesterol concentration was significantly positive for men with initial levels of serum cholesterol less than 250 mg/dl, significantly negative for men with initial levels of 250 mg/dl or higher and not significantly different from zero for all men together. The bias introduced by change in diet among hypercholesterolemic men differs importantly from bias due to unreliability of measurement and to interindividual differences in intrinsic level of serum cholesterol, because it can produce statistically significant but spurious correlations.