Specifications for procurement of vehicle detection systems have historically used performance comparisons with known accurate detectors, perhaps specifying one value of detection accuracy to represent several weather, lighting, and traffic conditions. In most cases, comparison with inductive loops has provided the necessary information by which to judge the performance of these detectors as either acceptable or unacceptable. However, differences in detection technologies are not adequately addressed with this method. A new concept for defining detection performance measures is proposed: it provides for some stochastic variation in sensor performance, within prescribed limits. In this case, video image vehicle detection systems (VIVDS) provide an example technology for applying this concept, but the concept is also appropriate for other technologies. Stochastic thresholds are defined that are consistent with field observation of three different VIVDS. Although these thresholds are quite large, it is believed that the combination of this new definition and a framework for defining stochastic performance will provide the basis for the detector industry to enhance its products.