On the tariff of penal sanctions, the police caution is considered as the least punitive, least criminalizing and most efficient way of handling 'less serious' cases at the pre-trial stage. Despite its humanitarian and managerial potential, however, cautioning gives great cause for concern. Research work spanning almost fifty years has systematically demonstrated deeply problematic aspects of the practice, and in the absence of any clear explanation, the discretionary power of the police in cautioning matters has served as a 'bucket theory' of observed discrepancies. This has led to a reformist politics which seeks out ever more sophisticated ways of regulating, standardizing and controlling cautioning discretion. Yet, the problems persist and, in some respects, are worsening. In the light of the explanatory and political bankruptcy of the received wisdom of police cautioning, this thesis seeks to develop an alternative perspective on the practice so that a different politics of reform can be formulated. The thesis takes as its starting point a questioning of the epistemological tenets of both conventional and critical cautioning knowledge, and from this analytical debris an alternative epistemological project is salvaged. This involves, first, a genealogical examination of cautioning practice through which an original question is posed - what kind of policework is this; and second, the development of a theoretical framework based on structurationist principles which re-conceptualizes cautioning discretion as structured and strategic action. Guided by these theoretical and conceptual commitments, an empirical study of cautioning practice in a Southern police force area is undertaken, and centres on a grounded analysis of the implementation of a 'new' prosecution and diversion initiative which aims to resolve cautioning issues through the application of 'rules'. Two key themes are explored through the lens of the implementation process. First, how is cautioning policy and practice constituted as policework and positioned in a common relationship with other forms of policing; and second, what kinds of policing values and ideologies are expressed by and reproduced in cautioning relations. What is proposed, then, is a theory of police cautioning which not only challenges conventional understandings of the practice but points to the need for a political agenda which moves beyond the simple formulation of a rules-discretion approach.