OBJECTIVE To describe the quantificational relationship between structures and cytotoxicitives of 33 flavonoe derivatives against HIV. METHODS 14 quantum-chemical descriptors have been calculated at B3LYP/6-31G ** level. A total of 5 important descriptors has been selected by stepwise regression analysis(SRA). These descriptors were: energy of the lowest occupied molecular orbit(ELUMO), molecular hardness(η), molecular polarizability(α), molecular dipole moment(μ) and net charge on the seventh carbon atom(Q 7). A leave-one-out cross-validation method was used to select the number of latent variables for the building of the QSAR models by principal component regression(PCR) and partial least square regression(PLS) method respectively. RESULTS The cross-validation squared correlation coefficients R cv2 were 0.70 for PCR and 0.74 for PLS, showing that both of these two models had good prediction ability and the PLS model was superior to the PCR model. CONCLUSION The QSAR model shows the effects of the five descriptors, and directs the investigation of the flavonoids with low toxicity.