Potential sources of variability in the measurement of solid oral drug products by near infrared reflectance spectroscopy were evaluated with statistical experimental design. Spectra were collected for two different tablet types according to the data collection and treatment parameters defined by the experimental design. Each tablet had three different dose-levels. Libraries were constructed using second-derivative spectra. Key figures-of-merit generated during internal and external library validation were used to calculate which parameters most strongly influence the library performance for dose-level discrimination. These responses and their corresponding experimental conditions were evaluated with the screening model in the JMP program. Segment value used for the second-derivative calculation was an an influential factor and had a complex effect. Orientation on the sampling platform also had an influential effect for embossed tablets. Collection of spectra over fewer days decreased variability within the library. More frequent reference spectrum collection improved the performance of libraries to a small degree. A larger sample population increased the range of spectral variability within a dose-level but apparently not the overall performance of the library. The number of scans averaged per spectrum was not an influential factor in this study. These results are summarized and used to recommend an approach to dose-level discrimination.