Attribute reduction is an important process in rough set theory.Finding minimum attribute reduction has been proven to help the user-oriented make better knowledge discovery in some cases.In this paper,an efficient minimum attribute reduction algorithm is proposed based on the multilevel evolutionary tree with self-adaptive subpopulations.A model of multilevel evolutionary tree with self-adaptive subpopulations is constructed,and interacting attribute sets are better decomposed into subsets by the self-adaptive mechanism of elitist populations.Moreover it can self-adapt the subpopulation sizes according to the historical performance record so that interacting attribute decision variables are captured into the same grouped subpopulation,which will be extended to better performance in both quality of solution and competitive computation complexity for minimum attribute reduction.The conducted experiments show the proposed algorithm is better on both efficiency and accuracy of minimum attribute reduction than some representative algorithms.Finally the proposed algorithm is applied to magnetic resonance image(MRI)segmentation,and its stronger applicability is further demonstrated by the effective and robust segmentation results. Attribute reduction is an important process in rough set theory. Finding minimum attribute reduction has been proven to help the user-oriented make better knowledge discovery in some cases. This paper, an efficient minimum attribute reduction algorithm is proposed based on the multilevel evolutionary tree with self-adaptive subpopulations. A model of multilevel evolutionary tree with self-adaptive subpopulations is constructed, and interacting attribute sets are better decomposed into subsets by the self-adaptive mechanism of elitist populations. More reviews it can self-adapt the subpopulation sizes according to the historical performance record so that interacting attribute decision variables are captured into the same grouped subpopulation, which will be extended to better performance both both quality of solution and competitive computation complexity for minimum attribute reduction. The demonstrated experiments show the proposed algorithm is better on both efficiency and accuracy of minimum att ribute reduction than some representative algorithms. Finaally the proposed algorithm is applied to magnetic resonance image (MRI) segmentation, and its stronger applicability is further demonstrated by the effective and robust segmentation results.