A strategy to improve accuracy of near-infrared (NIR) quantitative analysis of powder samples by alleviating the subsampling problem is demonstrated. The approach is to increase spectroscopic sampling volume for more accurate representation of its composition by filling the void space in a packed sample with liquid to reduce differences in refractive index between particle and unoccupied space. By this way, NIR radiation can propagate deeper into the packed powder due to decreased degree of reflection at the interfaces, leading to greater sampling volume. Perfluorohexane (PFH) was chosen as the space-filling liquid since it does not readily absorb NIR radiation and does not disintegrate sample particles due to its strong hydrophobicity. For evaluation, binary mixtures composed of ambroxol and lactose at different concentrations were prepared, and their diffuse reflectance spectra were collected in two ways: conventional direct powder measurement and PFH-incorporated powder measurement. The accuracy of ambroxol concentration determination improved in the PFH-incorporated measurement due to acquisition of more composition-representative spectra by the greater sampling volume. Also, the sample morphology did not change in PFH medium based on XRD analysis. In parallel, Monte Carlo simulation was executed to track NIR photons in the samples and explain the improved accuracy. Improvement in accuracy was also realized when analyzing real pharmaceutical samples containing an active pharmaceutical ingredient in granular form, demonstrating expandability of the proposed scheme.