Validation techniques are actuated by two primal problems in pattern recognition, which are selection of model and accuracy estimation. Selection of accuracy estimation method is very important in data mining. There are many accuracy estimation methods have been developed such as holdout. Cross validation, leave-one-out, bootstrap, random sampling, boosting and bagging. Cross-validation method with K-folds is used in studying large data sets, but it is also useful for small data sets in some aspects. In this paper the authors present a different form of cross validation with K-2 in which for each experiment. they used K-2 folds as training sets and the remaining two folds as test set. They repeat the experiments for 5 times and take average accuracy of the classifiers.