In order to realize accurate colorimetric characterization for a scanner, several polynomial regression models with different number of items and different conversion types were evaluated. The items of polynomial included 3, 5, 9, 11, 18 and 20, and the conversion types consisted of manifold transformation from scanner RGB space to standard CIEXYZ or CIELAB colorimetric space。An identical scanner was colorimetrically characterized with these polynomial models utilizing ANSI IT8.7/2 as color targets. Experimental results showed that the performance of these models were quite different and the linearization of scanner RGB signals improved the model accuracy when the items of polynomial were less than 11, while the differences disappeared and the linearization decreased the model accuracy along with the items more than 11. In this experiment, the polynomial conversion from raw scanner RGB to CIEXYZ with 20 items gave the best performance with the mean/maxmium color defferences of 0.832/2.988 CIELAB units.