According to a wide literature persons are not able to evaluate their own skills and knowledge, but the discussion on the nature, extension and determinants of this phenomenon is still open. This paper aims at proposing new empirical evidence on overconfidence and its determinants, trying to find out the possible effect of past performance on present optimism. I test my students' calibration and confidence in predicting their future results, comparing their expectations and real grades. My analysis allows showing the existence of overconfidence, its reduction in two following tests, and its non linear relation with students' capacities. Besides, I focus my attention on the effect of the grade my students got at the end of high school. This is used a proxy of their past experience and habit to get good or bad grades. Past success determined overconfidence. This idea is connected to the literature on heuristics and rule based perception.