There is increasing interest in clinical assessment of body composition, however uncertainty remains regarding the appropriate techniques. Dual energy X-ray absorptiometry (DXA) is often described as a gold standard, in view of its high precision (reproducibility). However, for the molecular model of body composition (dividing the body into water, fat, protein and mineral) the in vivo gold standard comprises the multi-component model, and recent comparisons of DXA against the four-component model have revealed wide limits of agreement between the techniques, as well as variable bias (inaccuracy) of DXA in relation to body size, gender and adiposity. DXA is therefore no gold standard for body composition, and has limitations both for casecontrol studies and for longitudinal investigations. Despite these limitations, DXA can remain valuable for clinical assessment providing that the sources of error are kept in mind. Inaccuracy in DXA arises because only in pixels containing no bone can soft tissue composition (fat versus lean) be assessed. In those pixels containing bone, of which the torso region contains many, soft tissue composition is estimated rather than measured.