The use of the independent contrast method in comparative tests is studied. It is assumed that: (i) the traits under investigation are subject to natural selection; (ii) closely related species are similar because they share many characteristics of their niche, inherited from a common ancestor; and (iii) the current adaptive significance of the traits is the focus of investigation. The main objection to the use of species values in this case is that third variables which are shared by closely related species confound the relationship between the focal traits. In this paper, I argue that third variables are largely not controlled by the contrast methods, which are designed to estimate correlated evolution. To the extent that third variables also show correlated evolution, the true relationship among the traits of interest will remain obscured. Although the independent contrast method does not resolve the influence of third traits it does, in principle, provide a greater resolution than the use of the species mean values. However, its validity depends on the applicability of an evolutionary model which has a substantial stochastic component. To illustrate the consequences of relaxing this assumption I consider an alternative model on an adaptive radiation, where species come to fill a fixed niche space. Under this model, the expected value for the contrast correlation differs from that for the species correlation. The two correlations differ because contrasts reflect the historical pattern of diversification among species, whereas the species values describe the present–day relationships among the species. If the latter is of interest, I suggest that assessing significance based on the species correlations can be justified, providing that attention is paid to the role of potentially confounding third traits. Often, differences between contrast and species correlations may be biologically informative, reflecting changes in correlations between traits as an adaptive radiation proceeds; contrasts may be particularly useful as a means of investigating past history, rather than current utility of traits.