Several hypotheses attempt to explain the latitudinal gradient of species diversity, but some basic aspects of the pattern remain insufficiently explored, including the effect of scales and the role of beta diversity. To explore such components of the latitudinal gradient, we tested the hypothesis of covariation, which states that the gradient of species diversity should show the same pattern regardless of the scale of analysis. The hypothesis implies that there should be no gradients of beta diversity, of regional range size within regions, and of the slope of the species-area curve. For the fauna of North American mammals, we found contrasting results for bats and non-volant species. We could reject the hypothesis of covariation for non-volant mammals, for which the number of species increases towards lower latitudes, but at different rates depending on the scale. Also, for this group, beta diversity is higher at lower latitudes, the regional range size within regions is smaller at lower latitudes, and z, the slope of the species-area relationship is higher at lower latitudes. Contrarily bats did not show significant deviations from the predictions of the hypothesis of covariation: at two different scales, species richness shows similar trends of increase at lower latitudes, and no gradient can be demonstrated for beta diversity, for regional range size, or for the slopes of the species-area curve. Our results show that the higher diversity of non-volant mammals in tropical areas of North America is a consequence of the increase in beta diversity and not of higher diversity at smaller scales. In contrast, the diversity of bats at both scales is higher at lower latitudes. These contrasting patterns suggest different causes for the latitudinal gradient of species diversity in the two groups that are ultimately determined by differences in the patterns of geographic distribution of the species.