This work deals with the prediction of the geographical origin of monovarietal virgin olive oil (cv. Koroneiki) samples from three regions of southern Greece, namely, Peloponnesus, Crete, and Zakynthos, and collected in five harvesting years (2001−2006). All samples were chemically analyzed by means of ¹H and ³¹P NMR spectroscopy and characterized according to their content in fatty acids, phenolics, diacylglycerols, total free sterols, free acidity, and iodine number. Biostatistical analysis showed that the fruiting pattern of the olive tree complicates the geographical separation of oil samples and the selection of significant chemical compounds. In this way the inclusion of the harvesting year improved the classification of samples, but increased the dimensionality of the data. Discriminant analysis showed that the geographical prediction at the level of three regions is very high (87%) and becomes (74%) when we pass to the thinner level of six sites (Chania, Sitia, and Heraklion in Crete; Lakonia and Messinia in Peloponnesus; Zakynthos). The use of classification and binary trees made possible the construction of a geographical prediction algorithm for unknown samples in a self-improvement fashion, which can be readily extended to other varieties and areas.