Because the bearing fault signal in various stages had some differences, and Wigner-Ville distribution with fixed kernel function could not adapt to different types of signals, pseudo WVD smoothened by variable kernel function was proposed to detect vehicle bearing fault. The characteristics function of the WVD was established and its crossterm interference was restrained by using the convergence of the characteristic function.On the basis,determination of different kernel functions was converted to optimization problems. Then the relation of collaborative optimization of kernel function shape and parameters were given.WVD with variable kernel was adopted to detect vehicle bearing fault.The results show that WVD with variable kernel has a stronger adaptability to signals and it can effectively restrain the oscillation of cross terms. The detection accuracy can reached more than 97% with micro growth of the calculation cost.