A method for constructing a distance histogram for nearest neighbor classification. A training sample is determined for each of two classes. For each defect, a distance is measured in the feature space between the defect and the training sample for each of the classes. All of the distances for a given defect are normalized by dividing each distance for the given defect by the sum of all of the distances for the given defect. A histogram is constructed by plotting a chart of the normalized distances versus the number of defects having the distances. A threshold bar is placed on the center of the histogram to construct a normal nearest neighbor classifier. The threshold bar can be adjusted to the left or to the right to construct a weighted nearest neighbor classifier.