This scientific report is based on results obtained within the framework of the Master internship of Julien Boyer (OPEX, University of Western Brittany (UBO), Brest, France). The study aimed to produce classification and quantification models to predict the cooking ability of yams genotypes using hyperspectral images. For this study the yams samples were harvested at CIRAD research station in Guadeloupe. The work was carried out with 10 contrasting varieties of yam presenting cooking ability ranging from very poor to very good. Samples were analyzed for their spectral profiles (HIS) and for their Pectin, starch, dry matter contents and texture properties (hardness) using specific SOPs developed in RTBfoods project. Different model based on average spectra and physico-chemical traits were developed in order to classify the genotypes according to their cooking ability and to quantify their pectin, starch and DM contents. The study demonstrated that there is a significant correlation between hardness, measured using a penetrometer and starch content (r = -0,80), pectin content (r = 0.52) and Dry matter content (r = 0.55). no significant correlation exists between starch, pectin and DM and cooking ability except for highly contrasted varieties (very bad vs very good). The models developed for predicting pectin, starch and texture content do not have sufficient performance for routine analysis, however these models can be improved by increasing the number of samples and the ranges of the constituents. The local PLS classification model is promising for the yam cooking ability classification with a classification error of 19%. To improve this model, it is recommended to collect more samples with more varied and better distributed cooking qualities. Indeed, there was within the dataset an over representation of middle-class variety and not enough good or bad classes. The model developed for dry matter quantification is efficient with an error of prediction RMSEP = 2.67%. T