This paper proposes a tool following function-based identification approach (TFFIA) for geometric errors of two rotary axes for one five-axis machine tool. It is comprehensive to identify all geometric errors of rotary errors. Firstly, synthetic error formulas of ballbar originate from the geometric error model of machine tools in order to consider the influences of 21 errors of three translational axes. It makes the approach more reasonable and precise. Secondly, the structures of three measurement patterns of TFFIA are described. Thirdly, in each pattern, errors of rotary axes affecting the accuracy of the sensitive direction are identified. As the result, the identification equations of all 20 errors coincide with the geometric properties of errors. Moreover, the impacts of setup errors of ballbar are eliminated with least square method to improve the precise of TFFIA. According to the structures of three patterns, only three installation of workpiece ball of ballbar are needed in the whole identification of two rotary axes to obtain the required ballbar readings. It greatly shortens the measurement time. Twenty geometric errors of two rotary axes are calculated with identification equations and ballbar readings. Finally, TFFIA is applied to a SmartCNC500 five-axis vertical machining center. The corresponding comparisons are proposed to verify the effectiveness and accuracy of TFFIA.