Abstract: Tropical cyclone (TC) track forecast errors have decreased considerably over the past several decades while there have been only modest intensity forecasting improvements, despite that the prediction of tropical cyclone intensity is an important and difficult task for not only scientists but also operational forecasters. In recent years, the best guidance of forecasting TC future intensity is the approach based on the statistical or statisticaldynamical models, such as the CLIPER (Climatology and Persistence) model. Since the multiple linear regression is the kernel technique in the CLIPER model, the statistical model set up by it could be inaccurate and instable when the predictors used to develop the equation are highly correlated with each other. In order to improve the CLIPER model through increasing its stability and decreasing its forecast error, the second generation of regression technique, so-called the Partial Least Square (PLS) regression, is introduced into the CLIPER model in this paper, and the current operational STI-CLIPER model is updated to the PLS-CLIPER model. The PLS-CLIPER and STI-CLIPER models are both applied to forecast TC intensities in the future 120 h over the western North Pacific from 2004 to 2007, and the intensity forecasting results made by them show that the PLS-CLIPER model is superior to the STI-CLIPER model within future 72 h. The prediction of TC intensity at the future 12 h from the PLS-CLIPER model is more accurate than the STI-CLIPER model through increasing the forecast tendency consistent rate by 10% and decreasing the mean absolute forecast error by 2 m/s. Particularly for the fastly intensified TCs whose increase of maximum surface sustained wind surpasses 10 m/s within 12 h, the forecast error from the former is around 4 m/s smaller than that from the latter. Besides these, the forecast model established by the PLSCLIPER model is more stable than the STICLIPER model, and the forecast error made by the former is independent on the TC intensity and its change, TC center position (latitude and longitude), and TC moving speed. Furthermore, the intensity forecasting accuracy is remarkably improved by the PLSCLIPER model for the TCs which include （1） the TC with the initial maximum surface sustained wind less than 50 m/s, （2） the TC that is in the intensified and sustained stages,（3） the TC occurring over the offshore of China, and （4） the TC moving westward or northwestward. These results indicate that, in the CLIPER model of choosing the same samples and potential physical predictors, more reasonable and more advancing technique of regression could lead to a more accurate forecast result. This work is a theoretical and algorithmic search for predicting the TC intensity under the framework of statistical-dynamical models in the future.