Particle swarm optimization (PSO) is a new heuristic algorithm which has been applied to many optimization problems successfully.Attribute reduction is a key studying point of the rough set theory,and it has been proven that computing minimal reduction of decision tables is a non-derterministic polynomial (NP)-hard problem.A new cooperative extended attribute reduction algorithm named Co-PSAR based on improved PSO is proposed,in which the cooperative evolutionary strategy with suitable fitness functions is involved to learn a good hypothesis for accelerating the optimization of searching minimal attribute reduction.Experiments on Benchmark functions and University of California,Irvine (UCI)data sets,compared with other algorithms,verify the superiority of the Co-PSAR algorithm in terms of the convergence speed,efficiency and accuracy for the attribute reduction.