A method is proposed to build a reliable and stable model for the quantitative determination of pharmaceutical tablets, where the active ingredient concentration of production batches does not vary enough to ensure a stable calibration model. Therefore, some synthetic tablets based on the same formula, but with an extended concentration range of the active ingredient, are produced and used to build a calibration model. This model is then adjusted by the use of production samples. The model used is based on multiple linear regression and feature selection. Validation by predicting new independent production samples shows that this procedure can be used for a satisfactory determination of the active ingredient.