Owing to the complicated geometric conditions and increasingly diversified driving environment, freeway curves have become a road section with frequent serious accidents. To ensure safe driving on curves, drivers must first perceive the movement condition and the vehicle's position, and accurately evaluate the characteristics of the road to make the right speed choice. In this process, the perception of speed plays a crucial role. The present study aims to investigate the driver's speed perception characteristics with respect to freeway curves to better understand the driver speed selection mechanism. We first construct six three-dimensional (3D) virtual highway models, which are consistent with the geometric lines and traffic engineering instruments of real freeways. A virtual simulation test is then conducted in a highly immersive environment. After the completion of the simulation experiment, we conduct a field verification experiment, test the actual driving speed and perceived speed of drivers in the same place as the simulation experiment, and verify the effectiveness of the simulation method. Finally, 25 3D curve models of four different types are constructed, and the influence of curve characteristics (e.g., curve radius, curve combination, tangent length between curves) on drivers' speed perception accuracy are tested on the simulation platform. The results show that the driver's perceived speed is lower than the real speed when driving on curves, and the tangent-to-spiral (TS) point is the section where the driver's speed is most underestimated. Radius was the most important factor affecting driver's speed perception, but the tangent length between curves also had a lesser influence. However, the curve combination had no effect on it. Our findings can help researchers and road designers understand the reasons for drivers' speed choice, thus promoting drivers' safety on freeway curves.