In order to solve the problem that the collected spectrum was influenced seriously by the operation method, improve the instrument sensitivity and enhance the differences between samples, reflectivity normalization method was used for data preprocessing in the present article. First, the reflect spectrum of tongue was collected, then the reflectivity and the normalized reflectivity of the samples were compared, and it was found that reflectivity normalization method can eliminate the impact brought by the collection from different distance. At the same time, the classification results between the reflectivity and the normalization reflectivity of the different samples were compared, and the results showed that the reflectivity normalization method can enhance differences between the samples and enhance model’s robustness and prediction ability of subsequent disease screening.