Abstract: For fast predicting the contents of lignin,pectin,total cellulose and total cell wall matter in flue-cured tobacco leaf by its main physical characteristics,210 samples of flue-cured tobacco leaves were collected from Zhaotong tobacco growing areas in 2010,8 physical characteristics(tensile strength,elongation,stem content,weight per leaf area,equilibrium moisture content,leaf length,leaf width,thickness) of and contents of lignin,pectin,total cellulose and total cell wall matter in each sample were determined.The prediction models for the contents of lignin,pectin and total cellulose and total cell wall matter were established by quadratic polynomial stepwise regression method,principal component regression analysis and BP neural network,and the models were tested with independent test data.The results indicated that the BP neural network model featured higher accuracy for the prediction of the contents of lignin,pectin,total cellulose,and total cell wall matter in leaves,the determination coefficient(R²) and root mean squared error(RMSE) for the predictive and measured values against the 1:1 line were 0.9900-0.9975 and 0.03-0.14,respectively.The prediction accuracy of polynomial stepwise regression model and principal component model was lower.The R² and RMSE of principal component model for the predictive and measured values against the 1:1 line were 0.15-0.37 and 0.025-0.990,respectively;those of quadratic polynomial stepwise regression model for the predictive and measured values against the 1:1 line were 0.004-0.093 and 0.21-1.60,respectively.