Attitude data can reveal culture’s secrets, but only if analysts acknowledge and transcend two problematic forms of heterogeneity. The first, relational heterogeneity, reflects the fact that the meaning of a response to a survey attitude question emerges from its relation to other attitudes: considered singly, the same response may mean different things to different respondents, depending upon the meanings with which they associate it. The second, population heterogeneity, a common problem in survey analysis, reflects the fact that attitudes may be related to one another in systematically different ways for different respondent subsamples. To overcome these challenges, we must use analytic methods that (a) focus on relations among attitude responses rather than on single responses and (b) partition survey samples into subsets based on patterns emergent from those relations. We use two such approaches, Latent Class Analysis and Relational Class Analysis, to examine Americans’ attitudes toward science and religion in the late 20th century, at the onset of a period of acute cultural contention between religious conservatives and secular liberals. Employing an unusually rich data set that enables us to take into account spiritualism (supernatural experience not sanctioned by formal religious institutions), as well as science and religion, we find that both LCA and RCA identify large subsets of respondents for whom science and religion are allied, rather than opposed. Moreover, RCA enables us to examine how the determinants of attitudes toward science, religion, and spiritualism are conditioned upon respondents’ construals of the relationships among them. This diversity of opinion among religious Americans and the presence of a previously overlooked religious constituency of science supporters, has important implications for science policy and science advocacy.