A new methodology for constructing calibration sets based on the use of laboratory samples encompassing the same variability sources as production samples was developed. The proposed methodology requires the use of no reference method in order to obtain reference values for the analyte; also, it provides more simple and robust calibration models than does the conventional methodology while retaining its predictive capacity. The procedure involves subjecting a set of laboratory samples spanning the desired API concentration range to a granulation treatment similar to that of the industrial process in order to obtain samples with the same physical variability as the production samples. The laboratory samples thus obtained are used to develop partial least squares (PLS1) calibration models in order to quantify the API in a pharmaceutical granulate. Based on the results obtained in this work, NIR spectroscopy is an effective alternative to the reference methods currently used for calibration. The proposed methodology requires no reference values to construct models; therefore, it can be regarded as an absolute analytical method. Also, it confirms the advantages of NIR spectroscopy as part of the process analytical technology (PAT) used by the pharmaceutical industry. A second aim has been the use of the multiplicative curve resolutionalternating least squares (MCR-ALS) algorithm to examine potential polymorphic transformations of the API during granulation.