Partial least squares (PLS) and principal component regression (PCR) are applied to lunar highland and mare soil data characterized by the Lunar Soil Characterization Consortium (LSCC) for prediction of abundance of lunar soil chemistry and minerals. The goal of this study is to develop a tool for predicting the abundance of chemical and mineral constituents of lunar soils from reflectance data. To determine the best model, the optimal number of PLS components is selected through the cross-validation. The results obtained from PLS and PCR using mean centered and normalized data are compared. The comparison indicates that using mean centered and normalized data has an insignificant effect on estimates of the soil chemical and mineral abundance. Both PLS and PCR perform well in predicting chemical abundance based on prediction errors. For most of minerals considered, one of PLS models always gives the best estimate of mineral abundances. PLS is preferred over PCR in estimation of lunar soil chemical mineral abundances because the first model uses fewer components than the second model. It is concluded that PLS has potential to be applied to lunar hyperspectral imagery (M3 mission) for compositional predictions.