A mobile sensor and sample-based algorithm (MOSES) to detect incidents on freeways is described. The proposed algorithm is based on statistical difference in the mean section travel time from two sets of probe vehicle samples before and during an incident. Unlike other incident detection algorithms, which operate at fixed time intervals, this sample-based algorithm is applied to detect an incident whenever a fixed sample of new probe vehicles has traversed a freeway section. The incident detection performance of MOSES at various sampling rates and probe vehicle percentages in the traffic stream has been tested on a set of data generated by a calibrated microscopic traffic simulation model. The results are compared with those of two of the most promising neural network incident detection models, which use input from stationary sensors and operate on a fixed time interval. When more than 50% of the vehicles are sampled as probes, MOSES can achieve a detection rate and false alarm rate comparable to that of the two neural network models but with faster mean time to detection and lower misclassification.