In high precision crankshaft following grinder, pre-compensation method is usually used to control crankshaft profile error. Obvious outliers may appear in measured profile error data if there are burrs or scratches in the ground part surface. If traditional Gaussian filter is still applied to process these data, compensation precision of profile error can be impacted obviously, and even scraps may be generated. In order to solve existing problems, Gaussian filter, Rk filter and robust Gaussian regression filter are used for closed profile error. The three filters are applied to analyze measured data. Comparing the results analyzed, it is obvious that the robust Gaussian regression filter has the strongest outlier removal effect of profile error in three filters. By changing the scale of outlier profile error, robust Gaussian regression filter can be validated with high reliability and adaptability. This filter method is integrated in pin-chasing grinder software. Outliers can be removed automatically. Compensation efficiency and following ground crankshaft profile error accuracy can be increased.