Abstract: Surface wave tomography is a very important tool to study shear wave velocity structures of crust and upper mantle. For surface wave phase or group velocity tomography we usually assume that surface waves propagate along great-circle paths. However, when the velocity structure has large variations, surface waves will propagate along off-great-circle paths, and therefore there may exist considerable errors in tomographic inversion results based on great-circle propagation of surface waves. We use the ray-tracing-based surface wave tomography method to analyze how off-great-circle propagation of surface waves influences the results of phase velocity tomography in western Sichuan. During the inversion, the fast marching method is used to track the ray paths of surface waves and the subspace inversion scheme is used for the inversion step. We use synthetic data from a theoretical model and short period ambient noise phase velocity dispersion data from the western Sichuan array for the analysis, and we compare the inversion results based on off-great-circle propagation and great-circle propagation of surface waves. Results of the synthetic tests show that the off-great-circle tomographic method recovers the anomaly better when the variation of the velocity structure is large.The inversion of real data from the western Sichuan array reveals that for the 6 s period data, the differences between the phase velocity tomographic results from the off-great-circle propagation and great-circle propagation methods are rather obvious, which are close to 0.2 km/s in the Sichuan basin. For the 10 s period data, differences between the two methods are much smaller, which are all less than 0.1 km/s.The main reason is due to that the 6 s period data are more sensitive to the complex shallow structure of upper crust, so the effect of off-great-circle propagation on the inversion results is more remarkable. Our results indicate that the off-great-circle propagation effect should be considered in the tomography if the observed phase velocities along different paths have large variations, for example, more than 10% of the average velocity, otherwise the inversion results of regions with large velocity anomalies might have considerable bias.