The reuse of structural timber plays a crucial role in promoting sustainability and circular economy principles. However, dismantling and reinstalling timber components often introduce modifications such as holes from fasteners (e.g., dowels, screws) or cavities from internal systems (e.g., piping, cabling), potentially affecting their mechanical performance. Assessing the residual load-bearing capacity of reclaimed timber is therefore essential, particularly for its application in seismic-resilient structures. This study introduces a visual grading-inspired methodology for classifying recycled timber based on hole characteristics. A stochastic numerical model is developed to simulate the bending strength reduction of beams with randomly distributed holes, generating extensive datasets for training predictive classification models. Machine learning and threshold-based classification approaches are employed to determine whether the reduction in bending strength exceeds a predefined 20% limit, which serves as a grading criterion. An experimental campaign on timber beams with controlled hole patterns supports model validation and establishes a practical classification threshold based on a single feature: the cumulative sum of hole diameters in specific beam regions. Results indicate that a simple conditional classification model achieves predictive performance comparable to machine learning techniques, yielding accuracy levels above 80% in distinguishing beams with bending strength reductions above or below 20%. This approach addresses existing gaps in current standards, providing an efficient and reliable method for assessing the viability of reclaimed structural timber, thus facilitating its safe integration into seismic-resistant construction.