The role of scale in ecology is widely recognized as being of vital importance for understanding ecological patterns and processes. The capercaillie (Tetrao urogallus) is a forest grouse species with large spatial requirements and highly specialized habitat preferences. Habitat models at the forest stand scale can only partly explain capercaillie occurrence, and some studies at the landscape scale have emphasized the role of large-scale effects. We hypothesized that both the ability of single variables and multivariate models to explain capercaillie occurrence would vary with the spatial scale of the analysis. To test this hypothesis, we varied the grain size of our analysis from 1 to just over 1100 hectares and built univariate and multivariate habitat suitability models for capercaillie in the Swiss Alps. The variance explained by the univariate models was found to vary among the predictors and with spatial scale. Within the multivariate models, the best single-scale model (using all predictor variables at the same scale) worked at a scale equivalent to a small annual home range. The multi-scale model, in which each predictor variable was entered at the scale at which it had performed best in the univariate model, did slightly better than the best single-scale model. Our results confirm that habitat variables should be included at different spatial scales when species-habitat relationships are investigated.