We evaluated bioelectrical impedance analysis (BIA) as a nonlethal means of predicting energy density and percent lipids for three fish species: Yellow perch Perca flavescens, walleye Sander vitreus, and lake whitefish Coregonus clupeaformis. Although models that combined BIA measures with fish wet mass provided strong predictions of total energy, total lipids, and total dry mass for whole fish, including BIA provided only slightly better predictions than using fish mass alone. Regression models that used BIA measures to directly predict the energy density or percent lipids of whole fish were generally better than those using body mass alone (based on Akaike's information criterion). However, the goodness of fit of models that used BIA measures varied widely across species and at best explained only slightly more than one-half the variation observed in fish energy density or percent lipids. Models that combined BIA measures with body mass for prediction had the strongest correlations between predicted and observed energy density or percent lipids for a validation group of fish, but there were significant biases in these predictions. For example, the models underestimated energy density and percent lipids for lipid-rich fish and overestimated energy density and percent lipids for lipid-poor fish. A comparison of observed versus predicted whole-fish energy densities and percent lipids demonstrated that models that incorporated BIA measures had lower maximum percent error than models without BIA measures in them, although the errors for the BIA models were still generally high (energy density: 15-18%; percent lipids: 82-89%). Considerable work is still required before BIA can provide reliable predictions of whole-fish energy density and percent lipids, including understanding how temperature, electrode placement, and the variation in lipid distribution within a fish affect BIA measures.