On the basis of elaborate analysis of traditional algorithm and relevant improved algorithms, an improved Information Gain (IG) algorithm based on word frequency information was proposed to solve the insufficient consideration of the frequency of features in traditional information gain feature selection algorithm. The improved algorithm modified parameters of the traditional IG algorithm, mainly from aspects of the frequency of features within category, distribution within category and the distribution among different categories, which can make full use of the frequency of features. The result of text categorization experiment compared with traditional IG algorithm and an improved IG algorithm of weighted indicates that the proposed algorithm has an obvious enhancement in accuracy of the text categorization.