Multiclass cancer classification remains a challenging task in the field of machine learning. We presented a comparative study of seven feature selection methods and evaluated their performance by six different types of classification methods. We applied it to the four multiclass cancer datasets. We demonstrated that feature selection is critical for multiclass cancer classification performance. We also demonstrated that an appropriate combination of feature selection techniques and classification methods makes it possible to achieve excellent performance on multiclass cancer classification task. Support vector machine method based on recursive feature elimination (SVM-RFE) feature selection algorithm combined with sequential minimal optimization algorithm for training support vector machines (SMO) classification method showed the best performance.