In cities, the main source of air pollutants is vehicle exhaust. It has become an urgent problem for many cities to explore the law of vehicle exhaust diffusion and actively seek for improvement measures. At present, there are many researches on the simulation of vehicle exhaust diffusion, but few on the modeling of diffusion process and volume visualization analysis. This paper integrates California line source dispersion model-4(CALINE4) model and 3DGIS to solve the problem of urban vehicle exhaust diffusion simulation and dynamic visualization. Firstly, a street-scale framework of vehicle exhaust diffusion simulation and visualization is established, and a continuous concentration field is established to express the diffusion processing. A unified data model is used in the semantic space to integrate the data of buildings, roads, trees and dynamic diffusion process data. Secondly, a graphics processing unit-based direct volume rendering method is used to visualize the dynamic diffusion data, and this algorithm is improved in real-time gradient calculation, texture sampling and time-based interpolation. Finally, this paper analyzes the vehicle exhaust diffusion of a road in the central business district area of Beijing. The experimental results show that the coupling application of 3DGIS and CALINE4 model can better reveal the spatial and temporal distribution characteristics of urban vehicle exhaust and concentration, which verifies the reliabless and efficiency of the proposed method.