The Principal Component Analysis (PCA) was applied to study the complex nonlinear relation between chemical component and smoke of tobacco. The cured leaves from five production regions were used as samples, projection and principal component pick- up were adopted to analyze the main factor influencing smoke components. The M5' model tree algo rithm was used to establish the subsection linear model of smoke components. The results showed that integration of the two methods could availably estimate the effect of chemical components on smoke components, and increase the forecast precision for smoke components.