The absorption band of water (1 334-1 539 nm) at different temperatures of Staphylococcus aureus suspension was detected and characterized to investigate the effects of changes in the temperature of samples and bacterial concentration on the molecular structure of water and to clarify the mechanism of detecting bacterial concentration. Different spectral pretreatment methods were compared to obtain an optimal one. Multi-level component analysis (MSCA) was used to establish a first-level model (quantitative spectra-temperature relationship,QSTR) and a second-level model (quantitative spectra-concentration relationship,QSCR) for the suspension of Staphylococcus aureus. The results showed that Savitzky-Golay convolution smoothing and continuous wavelet transform was the best pretreatment method. The QSTR can accurately characterize the temperature of samples,and its corrected correlation coefficient (RC) and corrected root mean square error (RMSEC) are 0.997 and 0.37℃,respectively. The wavelengths that contribute the most to temperature correspond to the water solvation shell,OH-(H2O)1,4 and superoxide,O2-(H2O)4,free water and free OH- (S0) or trapped water and water molecules with 1 hydrogen bond (S1). RP and RMSEP in the prediction set for the Staphylococcus aureus were 0.997 and 0.37℃. The QSCR can characterize the concentration of bacterial suspension with RC of 0.887 and RMSEC of 0.891 log CFU/mL for calibration. The wavelengths that contribute the most to the concentration of bacteria correspond to V3,H2O asymmetric stretching vibration,water molecules with 1,3 and 4 hydrogen bonds (S1,S3 and S4). For the prediction of Staphylococcus aureus samples based on the secondary level model,RP of 0.869 and RMSEP of 0.858 log CFU/mL were obtained. The models built in this study have good detection accuracy and adaptability,indicating that the structural characteristics of water play an important role in detecting the temperature and bacterial concentration of bacterial suspensions.