High dimensionality is a major problem in data mining. Microarray datasets available suffer from curse of dimensionality making data mining task very complex and time-consuming. This work addresses this problem of high dimensionality. Statistical techniques are employed for feature selection on two cancer datasets, i.e., colon and lymphoma. To measure the relevance of features selected, support vector machines, Naïve Bayes, K-nearest neighbor techniques are employed. The metrics used for comparison of performance are accuracy, error rate and F-measure. The results suggested that not all genes are necessary for correct classification. Only few genes get affected and are mutated for different cancers and identifying those genes can lead to accurate classification. The performance of various statistical feature selection techniques is compared for two cancer datasets namely colon and lymphoma. The relevance of selected features is tested by different classification techniques like SVM, Naïve Bayes, and KNN. The results showed that the best result for colon data set is 95.1% accuracy by the highest ranked 30 genes or features using Modified T-test with KNN classifier and 100% accuracy of lymphoma dataset with all of three techniques and all three classifiers with different selected subset.