Protein content in chestnuts is an important parameter for evaluating the fruit quality. In this work, the pilot application of near infrared (NIR) spectroscopy in measuring protein content of chestnuts was implemented. 182 samples were involved and 5 preprocessing methods were compared. PLS regression models developed from the spectra of intact nuts and peeled nuts were established separately. The results shown that, for the peeled chestnuts, the model based on spectra after First Derivative preprocessing performed better than other models with 0.9044 as the correlation coefficient (R) of calibration subset and 0.8029 as the correlation coefficient (R) of validation subset, respectively. For the intact chestnuts, the model established on the spectra after Second Derivative preprocessing got 0.8748 as the correlation coefficient (R) of calibration subset and 0.7324 as the correlation coefficient (R) of validation subset, respectively. The results indicated that NIR spectroscopy was feasibility to measure the protein content of chestnut rapidly and nondestructively.