Microarray technology is commonly used in the study of disease diagnosis using gene expression levels. The classification of cancer is a foremost area of research in the field of bioinformatics. Microarray technology enables the researcher to investigate the expression levels of thousands of genes in a single experiment and gives the measurements of their differential expression. The main drawback of gene expression data is that it contains thousands of genes and a very few samples. Feature or gene selection methods are used to successfully extract the genes that directly involved in the classification and to eliminate irrelevant genes. These methods considerably improve the classification accuracy. The proposed method applies bat algorithm (BA) for feature selection in tumour classification. Initially, the top-10 genes are selected by T-Statistics, signal-to-noise ratio (SNR) and F-Test. The classifier accuracy of k-nearest neighbour (kNN) technique is used as the fitness function for BA. The simulated results are demonstrated and analyzed with 10 different cancer gene expression dataset. For Lung Cancer Michigan and Lung Harvard2 datasets the proposed method gives 100% classification accuracy with minimum number of genes. For 5 other datasets, the proposed method gives more than 90% of classification accuracy. The results show the suitability of the proposed algorithm for feature selection in cancer classification.