This paper presents a prototype Dynamically Reconfigurable Vision (DRV) system. DRV is based on the intelligent, dynamic allocation of spatial and temporal resources in order to maximize system performance. Minimization of irrelevant data in the video processing chain reduces on- board processing requirements, power consumption, and payload, and increases the amount of relevant information that can be communicated over bandwidth-limited channels. Our DRV system employs a real-time reconfigurable CMOS image sensor which supports multiple variance-resolution, independently-configurable windows per exposure, operates in a snapshot capture mode to minimize smear, and outputs data through multiple video ports to minimize readout time. Multiple, time-correlated windows enable the vision system to better support multiple targeting functions concurrently, and achieve a maximum level of situational awareness. This imager is capable of reconfiguring itself in microseconds upon demand; frame-by-frame configurable parameters include frame rate, integration time, and parameters defining the position, shape, size, and resolution of each window. The system additionally features a small footprint and very low power, resulting from a CMOS implementation and on-chip signal processing using passive circuitry. The advantages of DRV over conventional imaging techniques are discussed, and the overall design and performance of our DRV sensor and camera are presented.