The destructive 1980 Campania-Lucania earthquake was a complex event, consisting of at least three subevents. While the main features of the first (0 s) and of the last (40 s) subevents are widely accepted, the second subevent (20 s) is still controversial. Fault geometries are here determined from levelling data, using a global minimization technique. Even if uplifts of southernmost benchmarks (the most sensitive to the 20 s subevent) are small, new proper statistical analysis allows us to draw a few robust conclusions. Our results agree with previously accepted models as regards the 0 and 40 s subevents. As regards the 20 s subevent, previous models always assumed a northeast-dipping fault, while levelling data show that it could have been equivalently caused by a rupture dipping either southwest or northeast. The statistical significance of the two models is about the same, and strike and dip of the two faults are consistent with the two focal planes previously obtained from teleseismic data. Aftershock distribution seems to favour the southwest-dipping fault model.