A total of 115 samples of camellia seed meal collected from Nanchang customs technical center are taken as research objects.The number of scanning times is 32, and the sample thickness is 4 mm. Standard normal variate (SNV), DG1 and SG9, which are the best pretreatment methods, are selected according to the influence of different pretreatment methods on the established near infrared models.The quantitative analysis models of ash content in camellia seed meal are established by partial least square method. The corrected correlation coefficient is 0.9698, the corrected root mean square error is 0.5236, the predicted correlation coefficient is 0.9575, and the predicted root mean square error is 0.6211.In order to verify the applicability of the models, the ash content of 15 camellia seed meal samples which are not involved in the establishment of the models are predicted. The predicted results are compared with the determination results of GB 5009.4-2016 by pair result t test. It is concluded that there is no significant difference between the results of this method and national standard method. This method will greatly improve detection speed and efficiency of the camellia seed meal quality, reduce the workload and the use of chemical reagents, which provides a technical basis for the realization of rapid and efficient quality classification and supervision of camellia seed meal.