Abstract: Absolute positioning accuracy can only be limitedly improved by the existing robot parameter identification, because the errors are not evenly distributed in the robot working space on account of the compliance of links and joints and other non-geometric factors. In view of these, a variable parameter error model is presented. Due to the coupling between the axis, space grid is presented to deal with the variable parameter error model in order to simplify the solution. Moreover, a kind of regular sampling point selection method is proposed. Futhurmore, the modified Levenberg-Marquardt iterative least-square is applied to the global convergence solution of the parameter error of each space grid. Finally, a kinematic parameter calibration experiment on the KUKA industrial robot is completed with the help of a laser track to demonstrate the effectiveness of the compensation. The result shows that the average absolute positioning accuracy of the robot can be improved from 0.901mm to 0.115 mm.