Diabetes mellitus is one kind of chronic diseases which seriously threaten human health. It is very important to diagnose in the early stage. With the development of diabetes, the structure and function of erythrocyte in the blood will change. So the peak position and peak height of erythrocyte fluorescence spectrum are different. These differences can be used to determine the status of diabetes. In the selection of the difference of spectral signal as the feature vector, the nonlinear degree of fluorescence spectrum can be used as the feature vector. In order to describe the nonlinearity of the fluorescence spectrum signal, the nonlinear degree of the signal is described with the delay vector variance (DVV) method. By using the method of iterative amplitude adjusted Fourier transformation (IAAFT) to generate surrogate data of raw data, the nonlinear characteristic of the raw data is determined by comparing the DVV of the original data and the surrogate data. The variance of the original data is the horizontal coordinates, and the variance of the surrogate data is the longitudinal coordinates, thus the DVV scatter plot is drawn. The DVV scatter plot of healthy rat erythrocyte fluorescence spectrum is almost coincident with its diagonal, which means the nonlinear degree of healthy rat erythrocyte fluorescence spectrum is lower. The DVV scatter plot of diabetic rat erythrocyte fluorescence spectrum deviates from its diagonal, which means the nonlinear degree of healthy rat erythrocyte fluorescence spectrum is higher, also the corresponding amino acid spectrum nonlinearity is deeper. Therefore, it is proposed that the nonlinear difference between the healthy and diabetic fluorescence spectrum can be used as a feature of early diabetes diagnosis.