K-means algorithm is one of common clustering algorithms,  but the cluster center initialization is a hard problem.  In this paper,  a hierarchical-based initialization approach is proposed for K-Means algorithm.  The general clustering problem is treated as weighted clustering problem,  the original data is sampled level by level to reduce the data amount.  Then clustering is carried out at each level by top-down.  The initial center of each level is mapped from the clustering center of upper level and this procedure is repeated until the original data level is reached.  As a result,  the initial center for the original data is obtained.  Both the experimental results on simulated data and real data show that the proposed method has high converging speed,  high quality of clustering and is insensitive to noise,  which is superior to some existing clustering algorithms.