The weigh-in-motion system is one of the important part of the bridge structure health system, which can provide abundant vehicle load monitoring data. On this basis, establishment of vehicle load model that can reflect the actual traffic conditions is of great significance for safety assessment and vehicle-induced fatigue analysis of bridge structures. This paper proposes a random traffic flow simulation method and process that considering different traffic states. In the process of vehicle load statistical analysis, the traffic states are divided by the vehicle flow intensity in each period. For different traffic states, the common single-peak probability distribution models and the multi-peak one are introduced to carry out probability fitting for vehicle weight, vehicle speed and vehicle spacing, and the optimal probability distribution of vehicle load parameters can be obtained by K-S test. The Monte-Carlo method is used to simulate random traffic flow, and then decomposed into random loading flow. Finally, the rationality of the vehicle load simulation method is verified based on the vehicle load monitoring data of an actual bridge. The results show that when the different traffic states are considered for random traffic flow simulation, the stress amplitudes and stress cycle times of the steel box girder lower flange at mid-span are close to the results induced by the actual traffic flow. By contrast, when the different traffic states are not distinguished for traffic flow simulation, the stress amplitude and the number of cycles obtained are significantly smaller than the actual, which would lead to risky results for the vehicle-induced fatigue analysis of the bridge structure.