Based on vehicle suspension system model, a Kalman filter was designed to estimate vibration velocity by measuring sprung mass acceleration and unsprung mass acceleration. The influence of process noise covariance matrix on vibration velocity estimation was analyzed. Then an adaptive Kalman filter based on predictive filter was discussed. The simulation results show that the designed Kalman filter can accurately estimate vehicle vibration velocity in real time. Inaccuracy of process noise covariance matrix has larger influence on body velocity estimation, or even makes the filter fail. The adaptive Kalman filter can compensate the estimation error caused by the unknown process noise covariance and obtain accurate vehicle vibration velocity.