The influence of order number and selection of color spaces on polynomial regression model was analyzed. Color scanner characterization method based on nonlinear polynomial regression model was studied. Experimental results showed that as the order number features increase, the characterizing precision on training results of the nonlinear polynomial regression increases and the precision of experimental results decline in the high order. Polynomial equation present ill, and its generalization ability drops; characterizing precision of the polynomial regression of RGB to CIE transformation is higher than RGB to XYZ. The results showed that polynomial regression model can satisfy the accuracy requirement of scanner characterization.