Neighborhood based classifiers are commonly used in the applications of pattern classification. However, in the implementation of neighborhood based classifiers, there always exist the problems of uncertainty. For example, when one use k-NN classifier, the parameter k should be determined, which can be big or small. Therefore, uncertainty problem occurs for the classification caused by the k value. Furthermore, for the nearest neighbor (NN) classifier, one can use the nearest neighbor or the nearest centroid of all the classes, so different classification results can be obtained. This is a type of uncertainty caused by the local and global information used, respectively. In this paper, we use theory of belief function to model and manage the two types of uncertainty above. Evidential reasoning based neighborhood classifiers are proposed. It can be experimentally verified that our proposed approach can deal efficiently with the uncertainty in neighborhood classifiers.