CCOS (computer-controlled optical surfacing) approach is widely used in different optical polishing processes for making ultra-precision components with high surface form accuracy requirements. Under the polishing mode using complex tool motions, middle-spatial-frequency errors (MSF, i.e., residual ripples) tend to be generated on the workpiece surface, which would deteriorate the functional performance of the parts. In this paper, a dynamical removal model (DRM) based on the spatial kinematic analysis and polishing material removal theory is proposed, to investigate the actual material removal characteristics in dual-axis wheel polishing (DAWP). The model is developed to calculate the overall material removal profile (low-spatial-frequency error, LSF) and the middle-spatial-frequency errors, simultaneously. The shape and the stability of the tool influence function (TIF) are firstly investigated. A series of straight line path polishing tests on BK7 glass are carried out to investigate the effects of process parameters on the generated MSF errors and also check the validity of the proposed DRM. Uniform polishing tests on 50 × 50 mm BK7 samples are conducted, under different polishing parameters, the power spectral density (PSD) characteristics of the surface are also analyzed. The results show that the proposed model can effectively predict the material removal profile and especially the ripple characteristics generated on the surface, which are not reflected when using the conventional convolution algorithm typically used in a CCOS process. When increasing the ratio of feed velocity to co-rotating speed, the wavelength and amplitude of the residual ripples increase, and the ratio of ripple amplitude to total material removal depth also increases. The level of residual ripples can be effectively restrained by the appropriate setting of the speed ratio (reducing the feed velocity or increasing the co-rotating speed).