Studies of diet in relation to disease raise methodological challenges considerably more complicated than in most epidemiological investigations. Diet is not a single exposure, but rather a complex set of many intercorrelated continuous variables. Moreover, these variables are likely to have non-linear relationships with disease and interact with each other. Many epidemiologists have questioned whether useful measurements of individual diets could be made within populations, due to homogeneity of food intake and imprecision of assessment methods. However, in several recent studies it has been possible to demonstrate reasonable levels of correlation between simple, structured food frequency questionnaires and detailed, weighed assessments of diet. This provides evidence documenting that between-person variation in diet does exist and that simple questionnaires are sufficiently accurate to measure these differences. For a limited set of nutrients, biochemical measurements provide an alternative assessment of exposure. It remains to be seen whether these methodologies can be reliably employed in case-control studies since even small biases due to the presence of illness, which may affect recall of diet or the levels of biochemical parameters, will seriously distort relationships between diet and disease.