In this dissertation, we construct an algorithmic framework for systematic tracking of moving sources in large-scale sensor networks. The tracking algorithms we developed generate the estimates of the tracking locations from fusion of space-time data by first fusing the data in space and subsequently by fusing the data in time. Fusion in space is performed by fusing current sensed data that is sufficiently high-quality from the sensor nodes to produce the current source location estimate. These location estimates are indexed as they become available and subsequently fused iteratively in time to produce tracking estimates. Both fusion in space and fusion in time are performed distributively over the ad-hoc sensor network by exploiting distributed algorithms of computation of averages. The distributed tracking algorithms are locally constructed at each participating sensor node exploiting only locally available sensor observations and local available network connectivity information. These algorithms we developed are also resource efficient, scalable, fault-tolerant and can readily adapt to local changes in network topologies. We present methods for optimizing and characterizing the performance of the algorithms as a function of the quality of the sensor measurements, the source dynamics, the sensor density and the network connectivity.