Based on near-infrared spectroscopy, four characteristic wavebands 4 277.63 - 4 3166.20, 4 887.06 - 4 941.07, 5 056.78 - 5 172.50 and 5 218.78 - 5 303.64 cm(-1), and two characteristic wavebands 4 902.49 - 4 817.64 and 4 740.49 - 4 107.91 cm(-1) were chosen to establish the partial least squares (PLS) regression model of water and adenosine in fermentation cordyceps powder, respectively. The prediction results of water and adenosine contents of the whole spectra PLS model were as follows: correlation coefficients (r) were 0.868 3 and 0.788 2, RMS error predictions (RMSEP) were 0.001 999 and 0.000 134, the remaining prediction deviations (RPD) were 1.974 4 and 1.640 7, respectively. However, using characteristic wavebands modeling can achieve a better performance with r of 0.869 1 and 0.829 0, RMSEP of 0.001 934 and 0.001 250, and RPD of 2.040 7 and 1.847 6 for water and adenosine respectively, and can largely improve calibration speed, providing the theoretical basis for the development of the testing instruments. So choosing the characteristic wavebands in this work to determine the water and adenosine in fermentation cordyceps powder is more effective.