Population-based research on diet, obesity and the metabolic syndrome is faced with accumulating evidence of biases that may profoundly affect results. One potential source of bias, which is often neglected in nutritional epidemiology, arises from self-selected study populations. Subjects who agree to participate in surveys may be at less risk of metabolic syndrome than those who refuse. Analogous to observations in adult populations, studies of schoolchildren have also yielded clear evidence of self-selection. Whether such selection patterns influence analytical results depends on how the biases relate to the dependent and independent variables being studied. Systematic dietary reporting error is another source of bias in studies of nutritional risk factors for disease. While obesity-related under-reporting bias is now well documented, less is known about whether specific foods and nutrients are disproportionately affected. However, two studies employing biomarkers for protein have suggested that obese subjects under-reported the proportion of energy from fat plus carbohydrate. This should alert epidemiologists to the possibility that a dual reporting bias may be present in studies of diet and disease: general under-reporting among obese subjects compounded by food-specific errors. In summary, biases due to self-selection and selective dietary under-reporting may produce consequences in epidemiological studies that are both unpredictable and complex. We conclude this review with recent findings involving dietary fat intake and regional adiposity in a population-based study of women. These preliminary results may have etiological relevance to the development of metabolic syndrome, but multiple biases of the type described previously may also be operating.