It is great significance to study the rapid detection for citrus greening because citrus greening is increasingly serious harmful for citrus fruit trees. In this paper, using Raman spectroscopy technology combined with partial least squares discriminant analysis (PLS-DA) method was used to explore the feasibility about rapid diagnosis citrus greening and the classification of disease. The Raman spectra of citrus leaves were obtained and leaves were divided into five types: slight greening, moderate greening, serious greening, nutrient deficiency and normal by common PCR. In the range of 715~1 639.5 cm-1, the three methods of first derivative, baseline correction and polynomial fitting were used to eliminate the spectral background to highlighted the characteristics peak of Raman spectra. Polynomial fitting were taken two times, three times and four times fitting in this method respectively, compared with the other two methods of first derivative and baseline correction for eliminated the spectral background. Combining with the least squares support vector machine (LS-SVM) and partial least squares discriminant analysis (PLS-DA), wedeveloped the discriminant models. By comparison, the effect of eliminated the spectral background using polynomial fitting was better than the other two methods. Especially the effect of PLS-DA model was taken two times fitting was the best The correlation coefficient of prediction (RP) was 0.98, while the root mean square error of prediction (RMSEP) was 0.67. The total misjudgment rate in the least was 0 andthe effect of LS-SVM model using the method of baseline correction was the worst, while the total misjudgment rate at maxium was 40%. The results showed that it was feasible to study the rapid identification of citrus greening by Raman spectroscopy technology, and a new approach to study the non-destructive detection of citrus greening was provided.