Abstract: The fingerprints of 181 flavor samples were extracted on the basis of GC/MS technology,models for categorizing the flavors into three categories(marked as A,B and C) were developed with principal component analysis.By setting 95% of confidence limit as controlling line and 99% of confidence limit as warning line,Hotelling T² statistical values were extracted.The developed models were validated with 50 flavor samples.The results showed that the flavors in Category A and B were correctly predicted,four unqualified flavors in Category C presented obvious quality fluctuation.This method possesses certain practical values for the quality control of flavors.