In order to study the ride comfort of semi-active vehicle, a detailed multi-body dynamic model of a passenger car was established by using SIMPACK software, and a seven-degree freedom mathematical model for the semi-active automotive system was also built. A model reference adaptive control based on neural network was designed for semi-active suspension system and worked out by means of Matlab/Simulink. The SIMPACK-Matlab co-simulation method was used to analyze the ride comfort of MR-damper semi-active suspension. The result showed that, comparing with passive suspension, when the car is running at 60km/h on grade C road, the neural network controller could reduce the mean square roots of the vertical acceleration, the roll angular acceleration and the pitch angular acceleration by 32.33%, 28.09% and 35.93%. When the speed of car achieved 120km/h, the mean square roots reduced by 41.56%, 18.52% and 22.97%. It is concluded that the model reference adaptive control based on neural network could reduce the vehicle body vibration and improve the ride comfort.