The spectra of both single sophora alopecuroide seed and single cassia tora l seed in natural state are determined by using Fourier transform near infrared diffusion spectroscopy. The viability of both seeds is distinguished by using distinguished partial least squares (DPLS). The result shows that the modeling based on the average spectrum at both side surfaces is better than that based on the single spectrum at one side surface. Both sophora alopecuroide seed model and cassia tora l seed model have an identification rate of 90% at least. This study provides a new method for distinguishing single viable seed rapidly, precisely and nondestructively.