Industrial serial robot has not only large geometric errors, but also nongeometric errors that can not be ignored, which limits its application in the field of high accuracy. A complete rigidflexible coupling position error model, including geometric and compliance errors was established, and a modified Levenberg-Marquardt algorithm (M-LMA) based on the predictive residual errors and the weighted recursive average filtering algorithm was used to identify the coupling error parameters. In order to improve the efficiency and reliability of the measurement process, an intelligent selection method of the measuring poses based on the linearly decreasing weight particle swarm optimization algorithm (LDW-PSOA) was proposed, which combined the external constraints of the detection features of the measuring equipment and the geometric characteristics of the endeffector. A local precise compensation method was proposed, which can be used simultaneously with the calibration or the global compensation respectively, and can also be applied directly alone. Meanwhile, a model optimization method based on the prediction accuracy and the number of parameters was proposed according to the characteristics of the robot and processing demands, and a multimode accuracy improvement strategy was formulated. Furthermore, the established models and the proposed algorithms were integrated into the development platform of Matlab to realize a GUI interface system. Finally, the experimental results showed that the proposed accuracy improvement strategy can not only achieve the performance of highprecision positioning of the robot in many ways, but also had an efficient and reliable measurement process.