A classification model using Vis-NIR spectroscopy (380–2000 nm) coupled with partial least square discriminant analysis (PLS-DA) was developed to segregate avocados in three classes, predefined by a warehouse using destructive FF tests performed on a small number of samples. This classification showed a satisfactory general accuracy of 62 %, with 100 % well-classified samples in Class 1, 20 % in Class 2 and 65 % in Class 3. To improve classification, a ripening index (RI) was developed, which combines FF and DMC. The discrimination ability in the three classes was tested using Wilk’s lambda, calculated as between-class variance to the total variance ratio. Results showed values of 0.648 for RI, 0.516 for FF and 0.038 for DMC. A regression model was subsequently developed using Partial Least Squares (PLS) regression to predict RI using Vis-NIR spectroscopy in an independent dataset. The PLS results were satisfactory with the whole spectrum wavelength range (380–2000 nm), with R2 of 0.62, SEP of 0.69 (), but presented a large bias value of 1.22 (). The same occurs in models developed in the wavelength range from 400 to 1100 nm, with R2 of 0.63, SEP of 0.68 () and bias of 1.03 (). This could be corrected using a bias and slope correction algorithm. Study of the correlation coefficients of the PLS regression models showed that the region 400–1100 nm has a huge influence in the model, which indicates the potential of using cost-effective short Vis-NIR spectrophotometers for RI prediction.