Near-infrared spectroscopy (NIR) was applied to the quantitative analysis of the concentration of alpha-mangostin (aM) in mangosteen pericarp powder (MP). The predicted results from the partial least squares chemometric method of various pretreatment data were compared to obtain the best calibration model. Two different types of containers (transparent capsules and glass vials) filled with the same samples were measured. For MP mixture in vials, the calibration model involving nine principal components (PC) could predict the amount of aM most accurately based on non-pretreatment spectral data. For MP mixture in capsules, the calibration model involving nine PC could predict the amount of aM most accurately based on first-derivative pretreatment spectra. The relationships of the calibration models for both samples had sufficiently linear plots. The standard error of cross-validation for the MP mixture in vials was lower and the R(2) values of validation were higher compared to the MP mixture in capsules. The equation for prediction of the concentration of aM in MP mixtures in vials is y = 0.9775x + 0.0425 with R(2) = 0.9950 and for those in capsules is y = 1.0264x + 0.0126 with R(2) = 0.9898. Both validation results indicated that the concentrations of aM in MP mixtures were predicted with sufficient accuracy and repeatability. NIR can be a useful tool for the quality control of herbal medicine in powder form without any sample preparation. The type and the shape of the container should be considered to obtain more accurate data.