Many transportation management agencies are using probe to collect average link travel time, which is significant for congestion management, traffic incidents detection, routing guidance and traveler information systems. A key issue of using probe technologies is to determine the probe vehicles sample size required to collect traffic information cost-effectively. This paper proposed a method based on random sampling theory to analyze the impact of probe vehicles sample size on accuracy of average link travel time estimation without considering the population distribution of traffic data. Finally, case study was performed with simulation and field data. The analysis result shows that (1) time interval has little affect on the accuracy of ALTT estimation while probe sample size is fixed; (2) when probe vehicles sample size reaches to certain level, estimated errors reduce gently with the increasing of the number of probe vehicles. The transportation management agencies may determine the optimal probe sample size according to the errors confidence interval estimated by the proposed method