Since the fifties, modelling as a supporting activity in policy making has been a longstanding endeavour in several research fields such as geography, transportation, economics, sociology and computer science. The attention on the role of modelling in planning has, during such a long period, been wavering: it was high in the fifties and sixties, even because of urbanization processes and post-war developmental issues, then it decreased and increased again because of the raising of environmental issues and sustainability. Over the decades the application of modelling in policy making has progressively evolved as a result of a number of changes which affected the whole modelling domain. These changes are manifold but they can be articulated under the overarching heading of the complexity approach. Talking about modelling in policy making however, the arguments for instantiating the complexity approach in policies are still in their infancy. In this paper an effort is made to identify some major aspects which may be helpful in contributing to the debate. Being more specific, the focus of the paper is on some contributions that the complexity approach and its related modelling techniques, such as agent-based simulations, can give to policy making. In this respect, a claim is made about the need for distinguishing two main perspectives on the changes taken to the fore by applying a complexity approach, although difficult to disentangle: i) the first perspective relates to the transformations which are occurring in the very role of the modelling activity as a consequence of both progress in information technology and epistemological and societal changes: these transformations are progressively breading various families of models, each of which play an autonomous role in helping policy-making in several ways; ii) the second perspective concerns the more radical shift that is occurring in the very conceptions of the modelled entity, i.e. the socio-economic system being enquired, as its cognitive ability is acknowledged as a constitutive component. While the issues raised by the former perspective have been increasingly addressed since the nineties (e.g., the development of decision support systems, GIS tools, scenario modelling, but also the integration of agent-based simulations with participatory methods, and all its methodological and practical consequences) and it will be briefly mentioned here, the latter is still largely unexplored and it is the main subject of the paper. The basic idea behind our work is that the step forward in recognising the relevancy of cognitive issues can give the opportunity of developing a more comprehensive conception of policies which goes beyond the idea of policies “rigidly” affecting the institutional settings in which socio-economic agents live. To say it differently, coping with the complexity of society and individuals can now mean also the possibility to enrich the comprehension of policies through their modelling as non static, rigid and exogenous institutional settings but as dynamic, co-evolving and interdependent elements, that is to say as “agents”. In this direction, discussion proceeds as follows. Firstly, attention is turned towards the building of a conceptual framework in which the notion of agency, i.e. the one based on reflexivity, pro-activity and interaction, is instantiated in relation to policy activities. This point is articulated according to two main dimensions: i) the time span according to which a policy is developed, that is to say the three main conceptual levels at which a policy is conventionally referred to, and namely the strategic, tactical and operational levels; ii) and the information requirement underpinning the developmental path of policy actions, i.e. recognition, guidance and capability. The joint consideration of these dimensions allows us to identify a few agency profiles of policy actions which are worth discussing. Such profiles will be developed along the presentation of an application of the suggested framework to a public policy at a regional level. A final claim in fact is made that whenever developing a simulation model for policy action, from the viewpoint here considered a range of possible advantages exist, which can mainly be of: i) improving the intelligibility of policy actions, that is to say getting a deeper understanding of a certain agency profile; ii) providing a linking among different profiles and policies, that is to say, for instance, connecting the tactical and the operational levels. The application of the approach to a public policy is made, as mentioned, on regional policies concerning road safety management. Such an application considers the exploitation of an agent-based simulation based upon the empirical data collected about both the phenomenon in society and the public policies for managing it. The case study, in particular, allows pointing out the value-added of multi level and multi time scale modelling, and the impact on policy management in the field. The work ends by summing up the arguments and by discussing the reaction firstly induced in regional policy makers by the just mentioned application and the possible developments of the work.