Near infrared diffuse reflectance spectroscopy (NIR-DRS) is a convenient and fast quantitative method for direct analysis of complex samples. The low sensitivity of NIR-DRS, however, is a problem for micro-analysis. A method for improving the sensitivity of NIR-DRS is proposed using selective adsorption. The method makes use of an adsorbent containing b-cyclodextrin to achieve the specific adsorption of bilirubin (BR). Preconcentration of BR from an aqueous solution with complex matrix was achieved through hydrophobic interaction and hydrogen bonding. Then NIR-DRS of the adsorbate was conducted and a partial least squares model was built for quantitative prediction. Both the preconcentration and the specific adsorption improve the analysis. The former enhances the spectral response by enrichment, and the latter makes the NIR response less affected by interferences such as metal ions, glucose etc. The results for the validation samples were a coefficient of determination (r 2 ) between the predicted and reference concentration of 0.98 with recoveries in the range 94.5-107.5% in a concentration range of 3.30-8.60 µg mL −1 . The concentration that can be quantitatively detected was as low as 2.60 µg mL −1 . Considering that the normal concentration range of BR in serum is 3.5-10.0 µg mL −1 , the method may provide a feasible tool for determination of BR in serum or other solution samples.