Multivariate models to classify Tuscan virgin olive oils by zone

In order to study and classify Tuscan virgin olive oils, 179 samples were c ollected. They were obtained from drupes harvested during the first half of November, from three different zones of the Region. The sampling was repea ted for 5 years. Fatty acids, phytol, aliphatic and triterpenic alcohols, t riterpenic dialcohols, sterols, squalene and tocopherols were analyzed. A s ubset of variables was considered. They were selected in a preceding work a s the most effective and reliable, from the univariate point of view. The a nalytical data were transformed (except for the cycloartenol) to compensate annual variations, the mean related to the East zone was subtracted from e ach value, within each year. Univariate three-class models were calculated and further variables discarded. Then multivariate three-zone models were e valuated, including phytol (that was always selected) and all the combinati ons of palmitic, palmitoleic and oleic acid, tetracosanol, cycloartenol and squalene. Models including from two to seven variables were studied. The b est model shows by-zone classification errors less than 40%, by-zone within -year classification errors that are less than 45% and a global classificat ion error equal to 30%. This model includes phytol, palmitic acid, tetracos anol and cycloartenol.