The question of the nature of our knowledge of society has recently been raised in an interesting form by Peter Winch in his monograph, The Idea of a Social Science, and debated in recent issues of Inquiry by A. R. Louch and Winch himself. In this paper I attempt to contribute to this discussion by attacking the problem of the nature of the empirical bases of social scientific knowledge, the main point in dispute between Winch and Louch. I try to construct an argument to show that in specifying the 'data' of social science, we have to introduce an element of 'interpretive understanding' which radically alters the meaning of the term 'empirical base' in social scientific contexts, thus supplementing Winch's argument in his reply to Louch. At the same time, my argument shows, I believe, that this view of the nature of social science does not lead to any arbitrary restrictions on the methods of research pursued by social scientists, as is sometimes imagined. What the argument leads to is the conclusion that our knowledge of society involves distinctive epistemological features that differentiate this kind of knowledge from the kind of knowledge we have in the natural sciences.