Radial variation and different sections are essential factors affecting the identification of wood by terahertz spectroscopy. To explore the influence of radial deviation and different facets on terahertz spectrum identification of timber, terahertz time-domain spectroscopy technology was used to obtain terahertz spectrum data of Cunninghamia lanceolata and Cryptomeria fortunei samples, a total of 600 sets of wood spectrum data, and then the terahertz spectrum were analyzed. A wood recognition model was established based on BP neural network. The accuracy of model prediction was compared. It was found that the terahertz spectra of wood samples were different due to radial variation and other sections. The prediction accuracy of the model with the same or different radial positions between the predicted sample and the modeling sample had little difference, and the highest accuracy was 96. 25%. When the expected sample and the modeled sample belong to the same and different sections, the difference in accuracy is massive. The results show that the technology based on terahertz time-domain spectroscopy can accurately realize wood recognition, radial variation has little effect on wood recognition, and different sections have a more significant impact on wood recognition.