Studying midazolam-induced amnesia offers an interesting approach to the organization of normal memory processes, since memory performance can be studied in the same subject in "on" and "off" drug conditions. The present study investigated the effect of midazolam on skill learning. The task and the experimental design we used also allowed us to assess the effect of midazolam on priming and explicit memory. Eighteen patients who underwent minor ear surgery and who were anaesthetized with intravenous midazolam, and 18 matched control subjects participated in a mirror reading task on 2 separate days. Patients were tested under midazolam on day 1 and without any medication on day 2. The mirror reading task was made of French words, some of which repeated across trial blocks, others being new. Patients under midazolam read new mirror written words faster with practice, which attested for intact skill learning. Moreover, they read repeated words faster than new words with practice, which was interpreted as reflecting intact priming abilities. These spared implicit memory capacities were observed along with severe explicit memory impairments. Learning for both new and repeated mirror written words on day 2 was similar in patients and in controls, which was interpreted as suggesting that the implicit learning that occurred on day 1 under midazolam was normal. The quality of skill learning, both in terms of speed of and lasting effect, was normal under midazolam in the task we used. In the context of the present task, midazolam offers an interesting, reversible model of amnesia.