During the process of building inversion model of water content based on NIR spectrum, it is the key issue to figure out whether the feature set of rock will change along with the differences of lithology or not. Aiming at this problem, firstly conducting a laboratory experiment on water absorbability of rock by intelligent test system of hydrologic action on deep soft rock, this paper measures near infrared spectrum of water-bearing rock in three different lithology at different times. For conglomerate, siltite and rammed soil, we collect 51 pieces, 106 pieces and 149 pieces correspondingly. Afterwards, first-order derivative method is adopted in dealing with pretreatment of original spectrum, avoiding the influence of environmental interference on the spectrum. Next, the geometric feature method is used to extract the spectral features and normalize it to eliminate the influence caused by different dimensions and change amplitudes. Then, the correlation degree between initial characteristic variables and that between initial characteristic variables and water content are analyzed, and the characteristic variables at two strong correlation bands are obtained by referring suppression threshold size and eliminating redundant features. At last, Maximal Information Coefficient （MIC）is used as a metric to compare and analyze the feature selection results on near infrared spectrum of water-bearing rock with different lithology, in order to evaluate the influence of lithology on the spectral characteristics of water-bearing rocks. The results are as shown below: （1）The near infrared spectra of conglomerate, siltstone and rammed soil have obvious absorption peaks near 1 400 and 1 900 nm, and with the change of water content, the absorption intensity becomes much more stronger, which shows a significant correlation with the size of the water content. （2）The maximum correlation coefficient between the characteristic variables of near infrared spectra of conglomerate, siltstone and rammed soil and their water content value shows that the correlation between the near infrared spectrum of rammed soil and water content is the strongest. （3）Each characteristic value of near infrared spectrum of different lithology has different correlation degree with water content, which shows that the peak height near 1400nm, the right shoulder width and the water content all have high correlation. However, the correlation will vary because of different lithology, with the right shoulder width and peak area in the vicinity of 1900nm having higher correlation coefficient with water content, and the correlation degree of right shoulder width being higher than that of peak area. （4）The characteristic variables of near infrared spectrum of water-bearing rock with different lithology are similar to the correlation of water content—peak height, right shoulder width and peak area are the three characteristics with the highest correlation degree.