In order to aidthe buckwheat breeding work, quantitative identification models for testing the content of protein and γ-aminobutyric acid (GABA) in perennial buckwheat leaves were built by near-infrared reflectance spectroscopy (NIRS) with quantitative partial least squares (QPLS). NIR spectra of 222 buckwheat samples were collected, and calibration models were established based on the spectra and chemical values. It was found the average, maximum and minimum protein contents of the samples were 164, 331 and 121 mg·g-1, respectively; the mean, maximum and minimum GABA contents of the samples were 2.489, 3.968 and 1.439 mg·g-1, respectively. Protein modeling results were as follows: when using different spectral regions, themean coefficient of determination (R2), standard error of calibration(SEC) and relative standard deviation(RSD) for the calibration set was 93.46%, 0.63 and 3.82% respectively, for the validation set, the mean R2, SEC and RSD was 91.77%, 0.88 and 5.28% respectively; when using different ratios of the modeling samples and testing samples, the R2, SEC and RSD for the calibration set was 93.55%, 0.63 and 3.82%, for the validation set, the mean R2, SEC and RSD was 92.18%, 0.87 and 5.20% respectively; when through second derivative (13) pretreatment, the wave number range of 4 000~9 000 cm-1 was appropriate for modeling (calibration sets∶validation set=4∶1), the R2, SEC and RSD for the calibration set was 93.57%, 0.55 and 3.38% respectively, for the validation set, the mean R2, SEC and RSD was 93.35%, 0.73 and 4.40% respectively. GABA modeling results were as follows: using different spectral regions, the mean R2, SEC and RSD for the calibration set was 86.28%, 0.21 and 8.30% respectively, for the validation set, the mean R2, SEC and RSD was 84.35%, 0.22 and 8.76% respectively; using different ratios of the modeling samples and testing samples, the mean R2, SEC and RSD for the calibration set was 88.51%, 0.20 and 8.04%, for the validation set, the mean R2, SEC and RSD was 86.80%, 0.21 and 8.40% respectively; no pretreatment, the wave number range of 4 000~10 000 cm-1 was appropriate for modeling (calibration sets∶validation set=4∶1), the R2, SEC and RSD for the calibration set was 93.28%, 0.15 and 6.10% respectively, for the validation set, the mean R2, SEC and RSD was 91.49%, 0.17 and 6.68% respectively. This study has demonstrated the feasibility and reliability of using NIRS to detect the content of protein and GABA in perennial buckwheat leaves.