In order to solve the missing item modeling of straightness error in the direction of ballbar error sensitivity for three-axis machine tools and error identification accuracy, taking the three-axis CNC machine tool as the research object, according to the product of exponentials and the three-axis CNC machine motion chain, a comprehensive error model was established combined with mixed-order Chebyshev pre-fitting mode. On the basis of the bi-orthogonal axis experiment, the comprehensive error coefficient model was solved by using the Moore-Penrose inverse matrix several times, so the 18 errors could be fully identified in the sensitive direction of the ballbar error, and it did not need to require decoupling in the non-moving axis, by which the identification accuracy and efficiency could be improved. The dual-orthogonal axis detection experiment and NC code compensation experiment were carried out on a three-axis CNC machine tool, the compensation errors of XY, XZ, YZ dual-orthogonal axis experiments were reduced by 82.02%, 91.63% and 70.6% after compensation, which verified the effectiveness of the proposed method. The residuals of the Chebyshev 〖JP2〗polynomial pre-fitting model before and after improvement were compared. The results show that the mixed-order Chebyshev polynomial pre-fitting model has higher accuracy.