Classifier performance is very important to give right decision above the data. Re-sampling methods, such as Jacknife, cross validation, are statistical methods to measure of classifier performance. The modified classifiers by using training set test with unused data in the training set, and introduce their performance results. The training and test sets are randomly selected to avoid the between correlation. But, this application creates that the classifiers are tried with test sets which are not compatible with training sets. If the classifier is based on interpolation, the classification risk of outlier samples is to increase. For that reason, the classifiers that are tested with re-sampling methods should have extrapolation properties. When the training and test sets are not overlapped, the success of classifiers can be decreased. The sets should be overlapped to decrease misclassification risk. The overlapping increases both correlation and performance. For that reason, a new deterministic re-sampling method is proposed as Give & Take. Each sample of data is separated into two groups according to their distance to center of class using distance metrics. Experimental studies show that, Give & Take gives better results than well-known validation methods.