Common shrews (Sorex araneus L.) were presented with two prey types at various densities in a depleting environment. Observed diet choice was compared to predictions of the classical optimal diet model and of "shrew-specific" simulations incorporating patch depletion. Two strategies were simulated: expansion of the diet from taking only profitable prey to taking both types, and fixed partial preference. The simulations predict partial preference over a narrow range of initial densities of profitable prey. However, within this range, energetic benefits are relatively insensitive to diet composition. Shrews preferred more profitable prey and were more selective when encounter rate with profitable prey was higher, broadly as predicted by all the models. Partial preference was observed, but neither of the mechanisms simulated was strongly supported by results for the shrews. Univariate and multivariate analyses were used to identify cues involved in selecting prey. Instantaneous measures of encounter rates and encounters per unit search distance were the best predictors of subsequent prey choice, but decisions appear to have been made on a probabilistic basis.