Feature selection (FS) is regarded as a combinatorial optimization problem that aims to search the optimal feature subset from large number of features. Evolutionary algorithm (EA) is an effective approach to solve FS problem. However, most EA-based FS methods tend to face an overwhelming number of feature combinations, which enlarges the solution space and makes it difficult to find the optimal feature subsets. As a result, a feature niching based differential evolution (FNDE) is proposed for FS on high-dimensional data. Specifically, the population is divided into several parts from the perspective of dimension using feature niching technique. In addition, the ranking-based selection strategy is used to choose the elite individuals based on their fitness rankings, so that high-quality solutions are preserved and poor ones eliminated. The experimental results show that the proposed FNDE achieves higher classification accuracy while selecting fewer number of features compared with other state-of-the-art methods on 8 benchmark high-dimensional datasets.