In this contribution, a Multi-Agent System architecture is proposed to deal with the management of spatially distributed heterogeneous nets of sensors, with specific focus on the problem of Pan-Tilt-Zoom or active cameras. The design of surveillance multi-sensor systems design implies undertaking to solve two related problems: data fusion and coordinated sensor-task management. Generally, proposed architectures for the coordinated operation of multiple sensors are based on the centralization of management decisions at the fusion centre. However, the existence of “intelligent” sensors capable of making decisions brings the possibility of conceiving alternative decentralized architectures. This problem could be approached by means of a Multi-Agent System (MAS). Specifically, this paper proposes a MAS architecture for automatically controlling sensors in video surveillance environments. In this case, the MAS fits perfectly in inherently distributed scenarios due to the area dimensions to be monitored. We deal specifically with Pan-Tilt-Zoom cameras to perform tasks depending on operator preferences, such as automated and distributed tracking for one or many targets, coordinated sweeps of areas, camera orientation to minimize overlapping areas, and so on. Both the architecture and an application example are described.