As the role of software increases, there have been several researches on understanding the characteristics of software and comparing such information. In this paper, to improve the performance of k-gram method for comparing two binary programs, we propose to apply approximate matching of k-grams. This method tries to improve the reliability of comparison results by considering partially matched k-grams in two programs. To evaluate the performance of the approximate matching of k-grams, we performed experiments on credibility and detection of similar programs in real world Java applications. In the experimental results, the proposed method can detect similar programs effectively compared to previous k-gram method. The results showed that the proposed method produced more reliable results for comparing programs. The approximate matching of k-grams is expected to be practically used to improve the previous k-gram method in analyzing similarity of programs.