One of the problems which may beset epidemiological studies is the difficulty of accurately measuring the dietary intakes of participants. Biomarkers of diet promise to provide a more accurate measure of dietary intake and a more objective one in that they are not reliant on the subject's memory. This review considers some issues of importance in epidemiology when information is obtained from biomarkers. The approach taken is to use examples both of normal dietary constituents and of contaminants in relation to a range of diet and health questions to illustrate these points. A brief overview of the role of sample collection, processing and storage is given including some generic recommendations for maximising the reliability of subsequent analytical data. Using the examples of phytoestrogens and iodine the question of whether biomarkers can accurately reflect the intake of the dietary constituents of interest at the population level or at the individual level is considered. The relationship of the biomarker to the natural history of the disease is exemplified using the role of folate in neural tube defects. Finally, intakes of vitamin D and heterocyclic amines are used to illustrate the integration of biomarkers into epidemiological studies of prostate and colorectal cancer, respectively. It is concluded that biomarkers may provide a more accurate and objective measure of diet than estimates of current or usual intake but that this approach also has limitations. A combination of methods will probably prove to be most valuable and this approach is being taken in current large prospective studies.