Microarray gene expression datasets comprise of a large number of genes in contrast to a small number of samples, thus having a high dimension of variables. Analysis of microarray data can lead us to many useful conclusions. In many microarray data analyses, selecting a small subset of genes which are of significance for a particular type of disease is an important issue but selection of such genes become difficult due to many irrelevant genes and noisy genes. The process of gene selection helps to extract the most informative genes, which consequently aid to build a robust prediction model using those genes. In this study, we employ a hybrid Chemical Reaction Optimization (CRO) based filter-wrapper methodology, which uses an information gain gene ranking heuristic to simultaneously extract informative gene subsets and build robust cancer classification models. The performance of the proposed method was tested on three benchmark gene expression datasets obtained from the Kent Ridge Biomedical datasets collection and the LIBSVM data repository. CRO results demonstrate its capability to select relevant genes with high confidence in comparison to the results reported earlier.