An idea of identifying unstable genes by integrative analysis of pair of different data has been proposed. This problem is modelled as a nonlinear integer programming problem. Three approximate methods have been proposed to work out the solution. An index is designed to measure and rank the instability magnitude of unstable gene. The experimental results on two lung cancer datasets from two research groups demonstrate the identified unstable genes have really unstable expression. The identified unstable genes can be used to improve the result of identifying differential expression genes and excluding these genes can effectively enhance the accuracy of microarray data classification. The findings suggest the proposed methods are effective and the unstable genes can provide valuable information for microarray data analysis.