Abstract-MircroRNAs (miRNAs) are one type of small noncoding RNAs and play important roles in regulating wide range of biological processes, such as stem cell maintenance, tissue development and cell metabolism. By modulating these key cellular processes, miRNAs can simultaneously regulate both oncogenes and tumor suppressor genes in cancer. With the development of high-throughput RNASeq technology, aberrant miRNA profiles and gene profiles have been detected in many cancers. And numerous studies indicate that vast miRNAs are involved in tumorigenesis and development of many cancers. They may act as either oncogenes or tumor suppressor genes in cancer and can be used as potential biomarkers for diagnosis, therapy and prognosis. To explore carcinogenic mechanism, it is critical to discover the aberrantly expressed miRNAs and their target genes. In this article, we propose an improved multiply-step approach to identify the target relationship between miRNA and gene pairs. An improved minimum redundancy maximum relevance (mRMR) feature selection method is used to get the differentially expressed miRNAs and genes. Then, we determine the significantly negative correlation between miRNA and gene pairs based on the expression level of selected miRNAs and genes in tumor and their corresponding adjacent normal tissues. To reduce the false positive rate of our approach, we check the relationship between miRNAs and their target genes by multiple miRNA target prediction databases.