Concerning the problem that the traditional Bee Colony Optimization (BCO) has good exploration but week exploitation performance, an exploitation enhanced BCO algorithm was proposed, and applied to data feature selection problem in order to improve the performance of the feature selection. Firstly, global weight was introduced into the food source, and was used to evaluate the importance of each food source to population, thus the randomness of exploitation was reduced; then, a recruiting method with two-step filtering was designed to improve the exploitation performance and keep diversity; at last, local weight was introduced into the food source to evaluate the correlation between the food source and class labels which were used in the feature selection model. Simulation experimental results show that the proposed method can improve the effect of the BCO and get a good performance in the feature selection problem, and the method outperforms Dissimilarity based Artificial Bee Colony (DisABC) and Feature Selection based on Bee Colony Optimization (BCOFS).