A predictive model of removal function in MRF has been established based on an analysis of MRF process. According to the residual surface error before and after finishing as well as the simulative residual surface error, this model can identify the efficiency coefficient of removal function with varieties of material and achieve exact prediction of removal function. Based on this model, the MRF process can be optimized by identifying the efficiency coefficient of removal function. A HIP SiC mirror which isin size had been polished nine times with this optimal MRF process. As a result, its residual surface error decreased from PV 0.72μm and RMS 0.108μm to PV 0.13μm and RMS 0.012μm, tested by sub-aperture stitching technique. The experiments indicate that the predictive model of removal function can optimize MRF process, improve the finishing determinacy and enhance the process flexibility so as to achieve high-precision and deterministic MRF of optical mirror.