On-machine measurement can be used for rea-time measurement during machining, which can effectively improve the machining quality and measurement efficiency of face gear. Considering the influence of the pre-travel error of face gear on-machine measurement on the tooth surface accuracy and the low tooth surface matching accuracy, an on-machine measurement strategy based on worm grinding wheel gear grinding machine, which takes various error influencing factors as input and the comprehensive pre-travel error of probe as output. A comprehensive pre-travel error prediction model based on PSO-CNN neural network is formulated to complete error compensation. A precise tooth surface matching method is proposed, and a six-parameter optimization model is established to match the measured tooth surface with the theoretical tooth surface to obtain the tooth surface error. Finally, the experimental results show that the accuracy of the left tooth surface is improved by 61.33%, and the accuracy of the right tooth surface is improved by 71.15%. The results of on-machine measurement are basically consistent with those of Klingeln Berg measuring instrument, which meet the requirements of on-machine measurement accuracy of face gear.