The adaptive management of vehicular tra ffic on roads is a key transportation application. Sensors are required to provide information describing the behaviour of traffic in the region to be monitored. There is scope for a low-budget, efficient and robust traffic monitoring system. The hypothesis is th a t an audio-based approach provides a highly economical and efficient solution to monitor road traffic. The main contributions of this thesis may be summarised as follows. In order to determine their behaviour over time, in d iv idual vehicles are successfully tracked with an efficient source localization technique based on acoustic information. The vehicle source location is determined by the inter-signal time delay of two cross-correlated microphones, known as the time delay of arrival (TDOA) localization method. A moving source model is derived from firs t principles to simulate the time-delay pattern due to changes in source location as a vehicle approaches and passes the array. Using the moving source model, two novel pattern extraction methods are developed to extract vehicle events and parameter values from the cross-correlation array. The first method minimizes the amount of cross-correlation data stored by extracting and tracking the path of predominant peaks, then comparing the path behaviour to the derived model to determine vehicle parameters. The second method draws on image processing techniques to search for regions or shapes of high correlation in the array that match the time-delay shape model of a passing vehicle. Each method was tested w ith real traffic data of 2,267 vehicles recorded at 5 locations under a range of conditions. The shape-matching approach yielded the highest accuracy of 93% for vehicle detection with a velocity tolerance of ± 19 km /h . The positive experimental results indicate th a t the preferred method is a viable, economical audio-based traffic monitoring sensor system.