As growing numbers of scholars become disaffected by the research traditions laid down by leadership psychology, there is a steady turn towards treating leadership as a discursive phenomenon. In response, leadership researchers are increasingly adopting interpretive and observational methods in the search for the practices of leadership in everyday life. This article suggests that while there are many advantages to an interest in discourse and action, there are also many subtle difficulties in making leadership observable and knowable in the field. Taking Louis Pondy's notion of leadership as a language-game as its starting point, this article argues that leadership studies as a discipline suffers from a persistent category mistake; a category mistake that some recent interpretive studies of leadership reveal, but inadvertently reproduce in the search for leadership's essential character. Instead, this article takes Pondy's thesis to its logical conclusion and outlines a programme of research that confronts this category mistake, whilst demonstrating the potential for, and limitations of, treating leadership as a language-game.