A trait‐taxonomic project was conducted in the North‐East of Italy. Three studies were run. In the first study, a comprehensive set of terms appropriate for personality description (adjectives, and adjectives that can be used as type‐nouns) was selected, and a modified version of the German selection procedure (Angleitner, Ostendorf and John, 1990) was adopted. In the second study, self‐ and other ratings were collected for a set of 314 trait‐descriptive adjectives. Self‐ and other ratings were factor analysed separately, and the factor solutions were compared in order to establish dimensions that were stable across the two data sets. In the third study, the pool of adjectives was reduced to 243 which were administered to a new sample, which also provided self‐descriptions on Goldberg's 40 bipolar scales. The resulting indigenous dimensions were compared with the canonical Big Five. Altogether, our findings suggest that the traditional Big Five are not reproduced when an emic taxonomy of Italian trait‐descriptive adjectives is developed. In fact, the three‐dimensional solution was the most robust in our studies (the Big Three), and not more than four factors remained stable across the data sets we analysed. © 1998 John Wiley & Sons, Ltd.