Heterogeneous sensor networks consisting of resource-constrained nodes as well as resource-intensive nodes equipped with high-bandwidth sensors offer significant advantages for developing large sensor networks for a diverse set of applications. Target tracking can benefit from such heterogeneous networks that support the use of sensors with different modalities. Such applications require tight time synchronization across the heterogeneous sensor network in order to improve both the estimation and real-time performance. In this paper we present a methodology for time synchronization in heterogeneous sensor networks. The synchronization methodology has been implemented as a network service and tested on an experimental testbed demonstrating an accuracy in the order of microseconds over a multi-hop network. In addition, we use the time synchronization method in a multi-modal tracking application for performing accurate sensor fusion of audio and video data collected from heterogeneous sensor nodes and we show that our method improves tracking performance.