Based on the attribute reduction algorithms computed with the equivalence relation in the neighborhood-rough set or fuzzy-rough set, the paper introduces the α information entropy for measuring fuzzy similarity relation, and proposes an attribute significance measure model based on it. Using the measure as heuristic information, a reduction algorithm for hybrid attribute is proposed.Comparing to other methods, experimental results show that the proposed method can select a few attributes but keep, even improve classification ability in several UCI datasets. In practice, by tuning α we can obtain multiple reduction results, and choose the optimum attribute reduction set according to the actual needs.