The idea of safety region estimation was introduced to state monitoring of rolling bearings, the safety region boundaries estimation and states identification of rolling bearings were carried out using a new method which is a combination of Local Mean Decomposition(LMD), principal component analysis (PCA) and least square support vector machine (LSSVM). Based on the collected vibration data of the rolling bearings under the four different states (normal, ball defect, inner race defect and outer race defect), the data was divided into a number of data pieces, and Product Functions (PFs) of each piece were got by LMD. And then, with the PFs, two statistical variables control limits as the state characteristics were calculated by PCA. The boundaries of safety region and identification results of the four states were obtained on two control limits data classification using two-classification LSSVM and multi-classification LSSVM respectively. Finally, the experiment results indicated that the accuracies of the safety region estimation and states identification are both satisfying, and shown that the LMD-PCA-LSSVM method is effective and feasible.