Abstract: This paper discusses the theoretical fundamentals of using auxiliary data in automatic classification of remote sensing images by the principle of pattern recognition:the simulation of the information space for visual interpretations, the influences of pattern dimensionality to the linear separability of pattern sample set, and separability measure to various kinds of patterns, i.e. OIF index and matrix trace criteria. The instance using auxiliary data for grass resources investigations in Xizha area of Northern Tibet has been given out to show the effects for improving the accuracy of automatic classification.