In recent years, the pharmaceutical industry has focused on a better understanding of the real-time manufacturing process with online measurements. Near infrared (NIR) spectroscopy is perhaps the most used and acceptable technique in such a highly regulated industry. However, one of the big challenges in using NIR systems for in-line manufacturing processes is the acquisition of measurements in real-time simultaneously at different locations. To this end, a series of different optical set-ups were investigated. Results are presented here using a multipoint NIR system based on a Fabry–Pérot Interferometer capable of recording simultaneously four independent spectra in four different points. NIR spectra of cellulose and sucrose granule mixtures were evaluated by the partial least-squares (PLS) regression method. By interpreting score values, loading of the principal components and the root mean square errors of cross-validation ( RMSECV), differences between two sensors composed of fibre optics and the effect of sampling (static or motion) were analysed. Sensors that allow large illumination and detection areas against small granule sizes showed better results in reducing the prediction errors ( RMSECV = 1.8%) than sensors with small illumination and detection areas against samples with large granule sizes ( RMSECV = 9.6%). These results are important to better understand errors associated with optic fibre probes and their configuration in multipoint NIR spectroscopy for monitoring the pharmaceutical manufacturing process.