Abstract: Geometric errors of rotary axes, which can be classified into position-independent geometric errors (PIGEs) and position-dependent geometric errors (PDGEs), are key factors affecting the machining accuracy of five-axis machine tools. Most existing identification methods only consider one type of error separately, making it difficult to accurately reflect the actual motion characteristics under the coupling effect of both types of errors. To address this issue, a stepwise identification method for geometric errors of rotary axes based on on-machine measurement was proposed to separate and identify PIGEs and PDGEs. Firstly, a geometric error model of the rotary axis is established based on dual quaternion theory, and the coupling influence law of the two types of errors on spatial pose is revealed. Secondly, a stepwise identification strategy is proposed. The actual axis of the rotary axis is identified by spatial circle fitting and the least squares method based on the measured coordinates of three precision spheres, and the PIGEs parameters are obtained. Subsequently, the PIGEs are substituted into the error model, a nonlinear least squares problem is constructed and the Powell algorithm is used to solve the six PDGEs. Finally, on-machine measurement experiments are carried out on a DMG DMU 60 five-axis machine tool, which verifies the effectiveness of the proposed method.