This paper proposes a data-driven approach for nonlinear process monitoring under the framework of locally weighted learning. Based on available process measurements, the locally weighted projection regression is used in the offline learning scheme to provide a series of locally weighted linear models, in which the algorithms of traditional projection to latent structures (PLS) and total PLS could be applied to establish improved test statistics suitable for complicated process monitoring. By using the weights of local models obtained from measurement learning, the developed test statistics are further online utilized to monitor potential abnormalities related or unrelated to process quality. The effectiveness of the proposed locally weighted total PLS monitoring approach is finally demonstrated by the comparisons with other relevant methods via simulations based on the wastewater treatment process benchmark under different abnormal conditions.