This study focusses on the ability of Near Infrared spectroscopy to predict chemical and Textural properties of Yam Tubers. To investigate it, numerous varieties of yams coming from a core collection (CIRAD-INRAe, Guadeloupe) and representing of the chemical and textural diversity were analyzed. A total of 174 samples were analyzed for their: DM, Starch, Protein and sugar contents in wet chemistry and for their texture properties (Hardness, Cohesiveness, Adhesiveness, Springiness and Extensibility) using a texturometer. The same samples were analyzed for their reflectance spectra in Near Infrared. Two replications of yam flour (dried) sample were scanned on a FOSS-NIR- Systems model 6500 scanning monochromator (FOSS-NIRSystems, Silver Spring, MD) with the autocup sampler. The whole sample set was divided in 3 data sets: learning set (N93), test set (N= 31) and external validation set (n= 41). Learning set was used in combination with the test set to set up the best fitting model in terms of error of prediction (SEP) and R². The external validation set was used to evaluate the performances of this model. The MPLS regression algorithm implemented in Winisi Software (Infrasoft International, Port Mathilda, USA) was used to develop the models. The performances of the different models ranged in terms of R² between 0.66 (extensibility) and 0.94 (sugar) and in terms of prediction error (estimated on external validation samples) SEP between 0.11 (for Cohesiveness) and 2.22 (Adhesiveness). The SEP were for DM = 1.58%, for sugar = 0.56%, for protein = 0.29% and for starch = 1.46%. This study demonstrated that it is possible to develop efficient predictive models based on NIRS spectra of Yam flour samples. These models are efficient for quantification of chemical parameters (starch, sugar, protein, DM). Models are less efficient, but promising, regarding textural parameters.