Abstract: Partial least square (PLS) regression was applied to the establishment of mathematical model in the NIRS analysis for initial boiling point of aviation kerosene. Emphasis was paid on the choice of methods of pretreatment of the near infra-red spectra and wave band of the model. As shown by the experimental results, best prediction effects were obtained by using the method of 1st derivation without window smoothing in the wave band combination of 764-960 nm and 1 000-1 020 nm. Values of r, SEC and SEP found were 0.910 2, 1.01 ℃ and 2.69 ℃ respectively, among which the values of r and SEC were found to be better than the value reported in literature (r=0.885 2, SEC=3.86 ℃). It was verified by t-test, that the results of prediction obtained by the proposed model showed higher accuracy.