Ground and air vehicles have distinctive acoustic signatures produced by their engines and/or propulsion mechanism. The structure of these signatures makes them amenable to classification by pattern recognition algorithms. There are substantial challenges in this process. Vehicle signatures are non-stationary by virtue of variations in engine RPM and maneuvers. Field sensors are also exposed to substantial amounts of noise and interference. We discuss the use of neural network techniques coupled with spatial tracking of the targets to carry out the target identification process with a high degree of accuracy. Generic classification is done with respect to the type of engine (number of cylinders) and specific classification is done for certain types of vehicles. This paper will discuss issues of neural network structure and training and ways to improve the reliability of the estimate through the integration of target tracking and classification algorithms.