[en] The aim of the present study was to develop and validate a near infrared method able to accurately determine a moisture content of pharmaceutical pellets ranging from 1% to 8% in order to check their moisture content conformity. A calibration and validation set were designed for the conception and evaluation of the method adequacy. An experimental protocol was then followed, involving two operators, independent production campaign batches and different temperatures for data acquisition. On the basis of this protocol, prediction models based on partial least squares (PLS) regression were then carried out. Conventional criteria such as the R(2), the root mean square errors of calibration and prediction (RMSEC and RMSEP) as well as the number of PLS factors enabled the selection of three preliminary models. However, such criteria did not clearly demonstrate the model's ability to give accurate predictions over the whole analyzed water content range. Consequently, a novel approach based on accuracy profiles which allow the selection of the most fitted model for purpose was used. According to this novel approach, the model using multiplicative scatter correction (MSC) pre-treatment was obviously the most suitable. Indeed, the resulting accuracy profile clearly showed that this model was able to determine moisture content over the range of 1-8% with a very acceptable accuracy. The present study confirmed that NIR spectroscopy could be used in the PAT concept as a non-invasive, non-destructive and fast technique for moisture content determination in pharmaceutical pellets. In addition, facing the limit of the classical and commonly used criteria, the use of accuracy profiles proved to be useful as a powerful decision tool to demonstrate the suitability of the proposed analytical method.