The 'multi-agency' approach to crime prevention has received considerable attention in recent years. This paper examines both theoretical and practical aspects of multi-agency work, in the light of research conducted over two years in a small number of neighbourhoods in inner London and a Lancashire town. Two contrasting forms of conventional understanding of the multi-agency approach—the 'benevolent as against the 'conspiratorial' thesis—are examined, and each found inadequate. Building on the evidence of research fieldwork, the paper strives for a more socially nuanced understanding which is alive to the complexities of locality-based crime prevention initiatives and of the power differentials running between different state agencies, as well as to the competing sectional interests within existing communities. Multi-agency strategies can undoubtedly have an impact on the lives of people within a locality, but these are not always the outcomes that are intended. This paper offers a preliminary attempt to conceptualise a perspective on the 'multi-agency' approach as it can be seen to operate in a specific number of localities, paying particular attention to the role of the local state. It is based on two years of fieldwork in four localities, three in inner London and one in a northern town, Milltown, which was undertaken as part of an ESRC funded project 'Crime, Community and the Inter-Agency Dimension'.1 The broad argument is that, at base, there is a fundamental set of conflicts between the state agencies we have focused on in our research, namely, the police, social services and the probation service.2 This structural conflict is either exaggerated or mediated by our other clear finding that there are structured power relations between the state agencies. In other words, in locally based crime prevention initiatives, some agencies are consistently more powerful than others. These two propositions form a framework for our arguments in the second part of the paper, which uses fieldwork examples to illustrate how, through multi-agency approaches to crime prevention and crime control, state 1 The paper was originally presented by Alice Sampson to the British Criminology Conference at the University of Sheffield, July 1987. It is based upon a collaborative research project between the University of Lancaster and Middlesex Polytechnic, funded by ESRC grant number EO 6250035. The research team wish to record their thanks to John Friend, who acted as management science consultant to the project. All the localities mentioned in the paper are identified by pseudonyms. 2 The focus of our research was originally state agencies within the criminal justice system, although this was necessarily broadened out during the course of the fieldwork, particularly in view of the prominent role of local authority housing departments in estate-based crime prevention initiatives. The research methodology involved interviews of agency personnel, together with extensive observational studies of police consultative committees, community forums, etc.