The surface topography errors of Micro-EDM are mainly composed of surface roughness, surface waveness and so on, which have influence on the workpiece’s functions and performances in various degrees. Thus, how to pick up these errors without distortion is quite important for evaluating the surface topography. It is well known that Wavelet Transform has good time-frequency localization properties, which is especially suitable for image processing. Therefore, in this paper, a wavelet method for surface roughness of Micro-EDM is presented, in which, the time of wavelet decomposition is determined by Pearson Chi-square goodness of fit test. Furthermore, the results of simulation and experiments show that the proposed methods can separate Micro-EDM’s surface roughness well, meanwhile, using the mean and variance of ideal surface as accuracy requirements of the reference surface could help the separation have a certain certainty and precision.