Abstract: By using a numerically controlled 2-D shifter,measurement data of PSD are obtained and then artificial neural networks are used for non-linearity correction of two dimensional PSD. When a light point moves in the detective area of PSD,the coordinates of the point at different positions are used as designed output of the networks and the output coordinates of the PSD are used as learning input of the networks,and a learning procedure is carried out. According to the ability of non-linear mapping of artificial neural networks,a linear relationship between input and output of the networks is set up after learning. Results show that learned networks can correct any non-linearity error in real time.