The objective was to implement a non-invasive classification system for green coffee beans by using near-infrared spectroscopy (NIR) and multivariate data analysis. For this, 4 types of coffee were analyzed, according to variety and geographical location. The samples were repeated 5 times. The observed NIR spectrum was absorbance in the range of 1100 and 2500 nm. In order to reduce the data, the analysis of main components was used by testing 24 classification models, from which the one that reached the highest level of precision was the Linear Support Vector Machine (SVM) algorithm, reaching 98.8%, achieving fairly satisfactory discrimination with values of PC1 (97.9%), PC2 (1.9%) and PC3 (0.1%), reaching a total cumulative variation of the contribution of the first 3 PCs of 99.9%. These values demonstrated that NIR spectroscopy is a valid alternative for classification by geographical origin and variety of green coffee beans.