Abstract: In order to meet the requirements for high accuracy in robot-assisted neurosurgery,this paper studies the positioning accuracy of our neurosurgical robot,NeuroMaster.Taking small distortions of the joint axes into consideration,the kinematic parameters are identified based on a revised Denavit-Hartenberg kinematic model.A multilevel feedforward neural network based on error backpropagation(BP) algorithm is used to compensate the transmission error of each joint.Posi-tioning accuracy of the robot is measured using an accurate three-dimensional measuring device.The result shows that the maximum value of the robot positioning accuracy is 1.63 millimeters,and the mean value is 1.04 millimeters,which are much better than the previous values.The robot can satisfy the requirements of most neurosurgical operations.