Abstract: To investigate nondestructive testing technology for starch content determination in tobacco seeds, near infrared spectra of 120 groups of tobacco seeds were analyzed within the wave length range of 4 000-9 000 cm^-1. A quantitative analysis model based on the near infrared spectroscopy for starch content in tobacco seeds was established by employing principal component regression (PCR) and verified by independent validation. The results showed that after the near infrared spectra were pre-processed with a standard normal variate (SNV) method combined with De-trending algorithm and smoothed with 19 points, the determination coefficient of the established model reached 99.80%, and the root mean square error of estimation (RMSEE) and standard error of prediction (SEP) were as low as 0.149 3 and 0.183 8, respectively. Meanwhile, the errors between the test values used in the validation set and the predicted values of the model were small.