To enhance quality breeding in Perilla frutescens, 250 lines of purple perillas collected from whole China were selected as material in the present study, combined with the technology of near infrared reflectance spectroscopy (NIRS) and partial least square method, NIRS Calibration Models for determination the content of oil. Palmitate (C16∶0), stearic acid (C18∶0), oleic acid (C18∶1), linoleic acid (C18∶2) and a-linolenic acid (C18∶3) were established, respectively. The results showed that, the coefficients of determination of all the models for calibration (RSQ1) were 0.98, 0.91, 0.92, 0.92, 0.85, 0.93, respectively. In addition, the cross validation correlation coefficient (1-VR) were 0.97, 0.89, 0.89, 0.91, 0.85 and 0.91, respectively while the external validation correlation coefficient (RSQ) were 0.98, 0.91, 0.89, 0.90, 0.80 and 0.89, respectively. All models above have proven credible as the low value for Calibration standard error (SEC) were 0.99, 0.21, 0.1, 0.94, 0.81, 0.92, respectively; Cross validation standard error (SECV) were 1.16, 0.23, 0.11, 1.05, 0.92, 1.02, respectively; and Standard error of prediction (SEP) were 0.97, 0.21, 0.11, 1.12, 0.99, 1.14, respectively, suggesting that these calibration models are accurate, feasible and highly efficient. The establishment of these NIRS Calibration Models can provide guidance in resource development and quality breeding of Perilla frutescens L and specifically are of great significance for breeding varieties with high oil content.