In this paper the autoregression correlation coefficient method is adopted to analyze the characteristics of the dry-season runoff of the rivers in the arid region of north western China. The paper first described the physical geographic elements affecting the dry-season runoff, and it is known that climatic elements and hydrological states are main ones which affect the dry-season runoff. The data used 10 calculate autoregression correlation coefficient from first step to tenth step are minimum monthly mean runoff and year minimum runof of the thirty representative stations in the north-west arid region. By analyzing the coefficients, the coefficients may be divided into three types, the first type is with low coefficients to any value of τ, the second with high coefficient ar T = l or 2 and the low values as τ>3, the third with the high values to any value of τ. However the type of the dry season runoff in the north-west arid region all belong to the first and second type, especially the first type. The autoregression model predicting and estimating the dry season runoff can be developed on the basis of the calculation and analysis of autoregression correlation coefficient.