This study describes the development of a quantitative structure-property relationship (QSPR) to predict the retention index (I) of volatile and semivolatile compounds identified in Arabica coffee samples from different geographical origin. The analytical method utilized rapid headspace solid-phase microextraction (HSSPME)-gas chromatography-time-of-flight mass spectrometry (GC-TOFMS) data measured in the divinylbenzene/carboxen/polydimethylsiloxane (DVB/CAR/PDMS) fiber. A total of 102 molecules were optimized with the PM6/ZDO level of theory using Gaussian 09 Rev D.01, in order to calculate 3006 molecular descriptors in the alvaDesc software. Initially, the number of descriptors was reduced to 1237 by means of the V-WSP unsupervised variable reduction to be submitted to the model development. The ordinary least squares were coupled to the genetic algorithms supervised variable subset selection to find the best three descriptors for the QSPR model. For model validation, the dataset was split into a training set (70%) and a test set (30%). The quality of the model was evaluated by means of the coefficient of determination (R²) and the root mean square error (RMSE) for the training set (R²_train=0.920 and RMSE_train=71.8) and the test set (R²_test=0.897 and RMSE_test=81.5). Other cross validation criteria were used such as leave-one-out (R²_loo=0.869 and RMSE_loo=91.7), leave-many-out (R²_lmo=0.876 and RMSE_lmo=93.1) and bootstrap (R²_loo=0.863 and RMSE_loo=94.2). The computational model accomplished the five principles defined by the OECD to be applicable as a tool for the identification of other volatile and semivolatile constituents in other coffee varieties.