Abstract: In this paper, the image signal and noise are described by models of mean-square differentiable two-dimensional homogeneous Gaussian random field. Under the conditions that signal and noise are independent and the mathematical expectation of noise is zero, the three important properties of the mean-square difference image matching system are proved. The analytical expressions of local accuracy for the meansquare difference matching system are derived. The theory and application basis are provided for accuracy analysis and evaluation of a system.