The analyses concern 94 cassava genotypes harvested in 2022 in Namulonge, Uganda. Genotypes came from NaCRRI Cassava breeding program. For this study 247 cassava roots were analysed; 3 uniformly sized non-necrotic roots per clone were sampled. Amylose quantification was achieved in NaCRRI physico-chemical laboratory in Namulonge. Near infrared spectra were scanned in NIR laboratory of NaCRRI in Namulonge. The protocol measurement follows the SOP protocol described for fresh grated cassava: https://doi.org/10.18167/agritrop/00669. Amylose content, expressed as % of starch, vary between 15,2% and 33%. On the basis of the 247 samples 2 data sets were constituted: one training set (n = 173) and one test set (n =74) for this 70% of the samples were picked randomly for the training set while the remaining 30% were kept as test set. Amylose content was calibrated using Modified Partial Least Squares Regression, for the spectral range: 400 nm - 2500nm, that is to say Visible and NIR regions. Calibrations were done on non-pretreated spectra and pretreated spectra using different pre-treatments. The best model was obtained with no pre-treatment applied to the spectra, the R² was 0,937 with an SECV equal to 0,801%. This model was applied to predict samples from the test set, the standard error of prediction (SEP) is equal to SEP = 1,023% and the R² for prediction is 0,915. The ratio performance to deviation RPD is equal to 2,947.The developed NIRs model for quantification of amylose content of cassava root presents an accuracy good enough to enable cassava selection based on amylose content expressed as % of dried starch. The error of the model is SEP = 1,02% which means that a predicted value could be defined with a confidence interval of +/- 2,04 % associated with 95 % of confidence. Furthermore, the database refers to 94 different genotypes representative of the variability of the starch properties, in particular for amylose content, thus the model is robust enough for amyl