The invention discloses a modeling method for grading quality of Xinjiang red dates in a near infrared range, which relates to modeling methods for grading quality of natural products in the near infrared range, and solves the problem that the prior randomly selected red date sample near infrared ranges differ greatly, main spectrum characteristic factors influencing spectrum change cannot be separated, and consequently, the grading modeling effect is poor. The modeling method provided by the invention comprises the following steps of: initially choosing appearance characteristics of samples, then, scanning the red date samples of same batch one by one; obtaining average spectrum of the batch of red date samples after performing proper spectrum treatment, and comparing the average spectrum with an original sample spectrum chart to screen the samples according to similarity so as to select the samples with greater spectrum difference, performing spectrum scanning again on the residual samples, and repeatedly adjusting and gradually subdividing the residual samples, until the spectrum difference of the batch of samples is in a needed accuracy range; and encoding the separated characteristic factors to create a sub-model. With the method, the characteristic factors of the Xinjiang red dates are successfully separated, and the sub-model is successfully created, so that the method provided by the invention is suitable for grading the quality of fruits, such as red dates, apples, pears and the like.