Models of near-infrared spectra under different resolutions, 1, 4, 16, 32 and 64 cm(-1), were studied with the mixed liquid samples of 4 components. The calibration models were developed by the method of partial least square and the validations of the models were carried out by the method of full cross. The value of target function was used to estimate the models performance. For the calibration models developed by the raw spectra, the target function values of benzene and benzaldehyde reached the max value with the resolution of 1 cm(-1), the target function value of toluene reached the max value with the resolution of 4 cm(-1), and the target function value of chlorobenzene reached the max value with the resolution of 16 cm(-1); for calibration models developed by the 1st derivative spectra, the target function values of the four components all reached the max value with the resolution of 1 cm(-1). The result suggested that, first of all, the resolution of the instrument will influence the quantitative analysis result. For the component with spectrum overlapped seriously, a higher resolution is good for the quantitative analysis, while for the analyte with a broad real band width, a lower resolution can be adopted in order to assure the signal to noise ratio. In addition, the influence of the resolution of the instrument is different for different components. Furthermore, the quantitative analysis result can be affected by both the SNR of the raw spectra and the band width of different components in the analyte, and a higher resolution is good for the quantitative model when the SNR of the spectra is assured.