System errors seriously influence on the leading and measuring precisions of a three-axis photoelectric tracking system. As least square method has lower precision

and global-function system error modification method could not properly be used in three-axis photoelectric tracking system when azimuth axis runs continuously

a novel neural network error modification method is presented in this paper. The system errors of three-axis photoelectric tracking system locate on a curve composed of three angle values

and the neural network can fit precisely complicated curves or curved faces. Analysis and simulation prove that BP neural network system error modification method can be used in three-axis photoelectric tracking system when azimuth axis runs continuously and can reduce system errors by 72%.