ABSTRACT Internal defects in the timber components of ancient architecture can reduce their effective cross-sectional areas, deteriorate their mechanical performance, and increase the risk of structural failure. To accurately and rapidly determine the extent of internal damage in components of ancient architecture, this study performed damage detection tests on Larix gmelinii (Larch) column specimens using stress wave nondestructive testing technology. An adaptive neuro-fuzzy inference system (ANFIS) model was developed using the cross-sectional wave velocity ratio after investigating the laws of defect identification through stress-wave testing. Stress-wave detection could identify the relative positional correlation of defect areas. However, the accuracy of the defect area detection was inconsistent (8%–98%). The accuracy of the ANFIS model exhibits a diminishing trend as both parameters rejection rate and influence range increase. However, the ANFIS outperformed the stress-wave detection technique and linear regression model with correlation coefficient and prediction accuracy of 0.99 and 88%, respectively. The root mean square error for the ANFIS was reduced by 79% and 70% compared to the stress wave detection method and linear regression model, respectively. Thus, the ANFIS model can precisely forecast areas of internal damage in timber components of historical edifices and can be used to evaluate the safety of ancient timber buildings.