Near-infrared spectroscopy (NIRS) is used in the pharmaceutical industry for monitoring drug content during the tablet manufacturing process. It is of critical importance to understand the effect of process factors on NIRS performance. Design of Experiments (DoE) methodology was applied in this work for the systematic study of the effects of compression pressure, precompression pressure and tableting speed on an average Euclidean distance (AED), which reflects spectral features of the tablets, and root mean-squared error of prediction (RMSEP) as key performance indicator of NIRS calibration models. Caffeine tablets were manufactured in 17 experimental runs in accordance with D-optimal design. Developed diffuse transmittance (DT) and diffuse reflectance (DR) calibration models were tested on five independent test sets to confirm the conclusions of the DoE. Compression pressure and tableting speed have shown significant effect on the studied responses in DT mode, whereas all three studied factors have shown a significant effect in DR mode. Significant factors were considered in the development of the global calibration models. The authors suggest further study of RMSEP and AED responses to draw reliable conclusions on the effects of tableting process factors. The global calibration model in DT mode has shown superior performance compared to DR mode.