In order to detect rice blast rapidly and accurately, chlorophyll fluorescence spectra of early rice blast were analyzed on leaf level, and the identification models of rice blast were established. Rice leaves were inoculated with rice pear spore first, and chlorophyll fluorescence spectra were achieved respectively at three stages of inoculation before (0h), gley period (48h) and disease spots early appearance (7d). Meanwhile, variation characteristics of chlorophyll fluorescence spectra at three stages were analyzed, Savitzky-Golay (SG) and the first derivative transform (FDT) were applied to reduce the noises and obtain the characteristics of chlorophyll fluorescence spectra. Then the method of Gaussian function fitting (GFF) was used to achieve the dimension reduction on spectral information, and multiple feature vectors of each band were extracted. Furthermore, the spectral data were divided into calibration set and validation set. Taking three stages of early disease as rice blast levels, and comparing four classic kernel function，support vector classification (SVC) models were established respectively with full bands feature vectors and composite bands feature vectors based on calibration set, and the models were tested with validation set. The results indicated that chlorophyll fluorescence spectra of blue green region, red and farred region were changed with the change of severity of early disease, GFF-SVC model with SG-FDT pretreatment for three stages disease had the highest classification accuracy rate, and the recognition results of different bands combination of primary spectrum, SG spectrum, SG-FDT spectrum were different for rice blast.