Bioelectrical impedance analysis (BIA) is the most commonly used body composition technique in published studies. Herein we review the theory and assumptions underlying the various BIA and bioelectrical impedance spectroscopy (BIS) models, because these assumptions may be invalidated in clinical populations. Single-frequency serial BIA and discrete multifrequency BIA may be of limited validity in populations other than healthy, young, euvolemic adults. Both models inaccurately predict total body water (TBW) and extracellular water (ECW) in populations with changes in trunk geometry or fluid compartmentalization, especially at the level of the individual. Single-frequency parallel BIA may predict body composition with greater accuracy than the serial model. Hand-to-hand and leg-to-leg BIA models do not accurately predict percent fat mass. BIS may predict ECW, but not TBW, more accurately than single-frequency BIA. Segmental BIS appears to be sensitive to fluid accumulation in the trunk. In general, bioelectrical impedance technology may be acceptable for determining body composition of groups and for monitoring changes in body composition within individuals over time. Use of the technology to make single measurements in individual patients, however, is not recommended. This has implications in clinical settings, in which measurement of individual patients is important.