Consensus algorithms for networked dynamic systems provide scalable algorithms for sensor fusion in sensor networks. This paper introduces a distributed filter that allows the nodes of a sensor network to track the average of n sensor measurements using an average consensus based distributed filter called consensus filter. This consensus filter plays a crucial role in solving a data fusion problem that allows implementation of a scheme for distributed Kalman filtering in sensor networks. The analysis of the convergence, noise propagation reduction, and ability to track fast signals are provided for consensus filters. As a byproduct, a novel critical phenomenon is found that relates the size of a sensor network to its tracking and sensor fusion capabilities. We characterize this performance limitation as a tracking uncertainty principle. This answers a fundamental question regarding how large a sensor network must be for effective sensor fusion. Moreover, regular networks emerge as efficient topologies for distributed fusion of noisy information. Though, arbitrary overlay networks can be used. Simulation results are provided that demonstrate the effectiveness of consensus filters for distributed sensor fusion.