Two testing methods, i.e., those using clear wood specimens (CS) and full-scale specimens (FS), are often used to determine the mechanical properties of lumber. The design index of wood in the traditional sawn and log timber structures is determined by the method of CS, and the result is different from that determined by the method of FS. However, there is no relevant research on the relationship or difference between these two methods. Therefore, 774 pieces of Chinese fir lumber graded by visual were selected, and then mechanical test, mathematical statistics, reliability analysis and mutual relationship were carried out or obtained to investigate the bending properties of both CS and FS. Results indicate that the influence of visual grade on the bending properties of FS is greater than that on the bending properties of CS. Lognormal distribution could provide a good fit for the tail probability distribution of bending strength for both CS and FS. The partial coefficients of flexural bearing capacity resistance that meet the requirements of target reliability are 1.34 for CS, and 1.11,1.15, 1.42 for Ⅰc, Ⅱc/Ⅲc and Ⅳc FS, respectively. The mean value of the bending elastic modulus of Ⅰc, Ⅱc/Ⅲc and Ⅳc FS is 1.16,1.13 and 1.08 times of the mean value of the bending elastic modulus of CS, and the design value of bending strength is 1.37,1.12 and 0.76 times of the design value of flexural strength of CS.