Correlations between habitat measures and animal distributions are not always applicable outside the study area that generated them. In such cases, the particularities of these correlations likely arise because only use of the local habitat has been quantified, rather than actual habitat selection, as the distribution models do not account for the behavior of animals in choice. The addition of covariates accounting for selection strategies could improve the precision and accuracy of correlative models of habitat use, but this conjecture has received little empirical attention. To evaluate this possibility, we re-assess previously developed habitat-use models for abundance of males of three grassland birds by explicitly including two measures of selection behavior: the “propensity to aggregate” and “propensity to use social information.” Habitat-use models for Nelson's Sharp-tailed Sparrow (Ammodramus nelsoni) were not improved by either behavioral variable. However, models for two other species, the Bobolink (Dolichonyx oryzivorus) and Savannah Sparrow (Passerculus sandwichensis), improved substantially through reduced prediction error (assessed with cross-validation) and were much more likely to be an appropriate model (by reducing the deviance of the fitted models). These results indicate that habitat-selection models can be an improvement over correlative habitat-use models. In our case, these improvements were limited to two species in which individuals use their conspecifics as cues of local habitat quality. However, numerous other measures of selection behavior can be included to improve upon certain habitat-use models, particularly when those models depart unexplainedly from optimality theory.