Principal component analysis (PCA) and its extensions have wide application in the area of quality assurance and process monitoring. One of the most important problems of the PCA-based method in chemical process monitoring is how to select the key components to improve the monitoring effectiveness. In PCA, the principal components (PCs) selected for each sample are the same and fixed, most of the criteria do not link the PCs’ selection with the fault detection, while the chosen PCs could be ideal for data modeling, they could not be for error detection. In order to preserve most important information, we proposed a method of weighted cumulative percent contribution (WCPC) criterion for PC selection. First, the influence of each latent variable (LV) in each sample’s T 2 statistic is calculated. Then assess LV’s significance for fault identification based on its impact on T 2 . After sorting contributions of LVs in descending order, the first few LVs with a CPC value above a threshold are chosen using the CPC criterion. Furthermore, each LV was weighted according to how frequently each LV was selected in all samples, LVs that appear more frequently have higher weights. The weight coefficient contains previous information. And the effectiveness and advantages of the proposed method are validated with Tennessee Eastman (TE) process and continuous stirred tank reactor (CSTR).