Cross-sectional and case-control studies are efficient and inexpensive approaches to the study of osteoporosis; their validity depends primarily on how the subjects are sampled. Retrospective studies of fractures may be biased if measurements change as a result of the fracture. Prospective cohort studies are less prone to such biases but are expensive. Randomized blinded trials are ideal for assessing studying single factors but trials are sometimes flawed by avoidable problems. Patient reports about risk factors are imperfect but sufficiently accurate to detect most strong associations between risk factors and osteoporosis. Noise in the measurement of risk factors will cause studies to underestimate the effect of these factors. About one-sixth of recalls of falls and fractures are incorrect. Self-report of vertebral fractures is unreliable and biased and there is controversy about how to define vertebral fractures on lateral X-ray film. Clinical and epidemiological research in osteoporosis would be advanced by greater care in the design of studies and methods of measurement.