Abstract: In this paper, an algorithm based on tabu search for selecting an optimal subset from original large-scale feature set is presented. The role and effect of the parameters in tabu search, such as the tabu list length, the neighbor size and the number of candidate solutions are analyzed. For two forms of feature selection problem, tabu search is compared with classic algorithms ,such as sequential and generalized sequential methods ,branch and bound methods, plus l and take away r method, etc. ,and other methods proposed recently, such as genetic algorithm and sequential floating forward(backward) search methods. The experimental results have shown that tabu search has good performance in both the quality of obtained feature subset and computation efficiency.