A geometric error model of a kind of blade computer numberical control (CNC) grinder is established based on the theory of multibody system and coordinate transformation method. The objective function for optimizationin in accuracy distribution was built on the basis of it. Then MATLAB genetic algorithm was used to optimize all the error parameters. Most of the optimized experiential design variables were amplified in different extends, which reduced the manufacturing cost of the grinder. The optimized values are all satisfied with the predefined processing accuracy, which well balances the processing accuracy and the economic performance.