Many studies have demonstrated that a teacher has a substantial impact on how much pupils learn. However, research on the long-term effects of teachers is scarce, and the results obtained so far are inconsistent. Some studies have concluded that teacher effects are relatively permanent, while others claim that they fade away more or less rapidly over time. The present study was conducted at the elementary-school level in France. Multilevel random crossed models were used to assess teacher effects over a period of two school years. The assessments showed that the pupils' test scores in French and mathematics after two years were linked to the second-year teachers but not to the first-year ones. These results suggest that teacher effects on pupils' learning gains wear off quickly over time. In other words, 'immediate' teacher effects on learning do seem to exist, but not long-term ones. The merits of multilevel random crossed models in this research field are presented, and the potential limitations of the results obtained are discussed.