The purpose of our research was to investigate efficient procedures for generating multivariate prediction vectors for quantitative chemical analysis of solid dosage forms using terahertz pulse imaging (TPI) reflection spectroscopy. A set of calibration development and validation tablet samples was created following a ternary mixture of anhydrous theophylline, lactose monohydrate, and microcrystalline cellulose (MCC). Spectral images of one side of each tablet were acquired over the range of 8 cm−1 to 60 cm−1. Calibration models were generated by partial least-squares (PLS) type II regression of the TPI spectra and by generating a pure-component projection (PCP) basis set using net analyte signal (NAS) processing. Following generation of the calibration vectors, the performance of both methods at predicting the concentration of theophylline, lactose, and MCC was compared using the validation spectra and by generating chemical images from samples with known composition patterns. Sensitivity was observed for the PLS calibration over the range of all constituents for both the calibration and the validation datasets; however, some of the calibration statistics indicate that PLS overfits the spectra. Multicomponent prediction images verified the spatial and composition fidelity of the system. The NAS-PCP calibration procedure yielded accurate linear predictions of theophylline and lactose, whereas the results for MCC prediction were poor. The poor sensitivity for MCC is assumed to be related to the relative lack of phonon absorption bands, which concurs with the characterization of MCC as being semi-crystalline. The results of this study demonstrate the use of TPI reflection spectroscopy and efficient NAS-PCP for the quantitative analysis of crystalline pharmaceutical materials.