To evaluate if the extraction criterion was correct and if the number of spectra extracted from each potato tuber could be reduced, PLS models were made extracting fewer spectra from each sample. First, a model was made using only the three spectra from the middle of each sample. This model had a significantly (p¼0.036) lower prediction performance (R2¼0.89 and RMSECV¼1.25) than the one using all spectra from each potato tuber. Secondly, two spectra from each end of the potato tuber were removed to see if removal of potential interference from the conveyor belt and reduced light scattering effects caused by the curvature of the potato tuber would improve the prediction performance. This also gave a model with slightly lower, but not significantly lower (p¼0.926), prediction performance (R2¼0.91 and RMSECV¼1.11) compared to the model using all spectra. This may also indicate that the threshold used to distinguish between the sample and conveyor belt worked as intended. This shows that a further reduction in the number of extracted spectra by removing spectra from each end of the sample will remove information rather than noise.