K-means algorithm is a frequently-used methods of partition clustering. However, it greatly depends on the initial values and converges to local minimum. In K-harmonic means clustering, harmonic means fuction which apply distance from the data point to all clustering centers is used to solves the problem that clustering result is sensitive to the initial valve instead of the minimum distance. Although the problem above is solved, the problem converged to local minimum is still existed. In order to obtain a glonal optimal solution, in this paper, a new algorithm called K-harmonic means clustering algorithm with simulated annealing was proposed. This alhorithm is introduced into simulated annealing to solve the the problems of local minimum. Then the algorithm was used to analyse IRIS dataset and get a conclution that the new algorithm get a glonal optimal solution and reached a desired effect.