An approach for recognition of transgenic soybeans was proposed based on spectral analysis in the terahertz (THz) range combing with Principle Component Analysis (PCA) and Back Propagation Neural (BPN) network. Eight principal component factors, whose accumulated variance reached 97.582%, were extracted from the original spectra data and then fed as inputs into the BPN network model. The utilization of the dimension-reduced data in training the network model can recognize the validation set accurately. The nondestructive testing of transgenic soybeans could be achieved by using THz spectroscopy, which could be widely applied in agricultural security areas.