In order to improve the accuracy of nondestructive testing methods to predict the bending strength of ancient building wood, multi-point nondestructive test was carried out on ancient Tibetan Populus cathayana bending specimens with stress wave detector and resistograph detector. By comparing the bending strength of the specimens, the correlation between nondestructive testing index and the bending strength was analyzed. The influence of measur? ing method, location and number of measuring points on the correlation coefficient, as well as the optimal combina? tion of independent variables in the multivariate model were discussed. The results indicate that the correlation coeffi? cient between nondestructive index and bending strength increases with the increasing of the number of measuring points; the stepwise regression method can quickly determine the optimal combination of independent variables in the multivariate model, while avoiding the colinearity of independent variables. Compared with the univariate linear model, multivariate linear model determined by the optimal combination of independent variables can increase the correlation coefficients and adjust R square between nondestructive index and the bending strength by 20.54% and 42.92%, respectively, which proves that the multi-point nondestructive testing method can effectively improve theprediction accuracy of bending strength of ancient building wood.