The aim of this study was to develop a stochastic model for predicting the bending strength distribution of glued-laminated timber (GLT). The developed model required the localized modulus of elasticity (MOE) and tensile strengths of laminae as input properties. The tensile strength was estimated using a regression model based on the localized MOEs and knot area ratios (KAR) which were experimentally measured for lamina grades samples. The localized MOE was obtained using a machine stress-rated grader, and the localized KAR was determined using an image-processing system. The bending strength distributions in four types of GLTs were simulated using the developed GLT beam model; these four types included: (1) GLT beams without finger joints; (2) GLT beams with finger joints; (3) GLT beams with different lamina sizes; and (4) GLT beams with different combinations of lamina grades. The simulated bending strength distributions were compared with actual test data of 2.4 and 4.8 m-long GLTs. The Kolmogorov–Smirnov goodness-of-fit tests showed that all of the simulated bending strength distributions agreed well with the test data. Especially, good agreement was shown in the fifth percentile point estimate of bending strength with the difference of approximately 1%.