[en] Several studies using the mirror-reading paradigm have shown that procedural learning and repetition priming may be preserved in the early stages of Alzheimer's disease (AD) (e.g., Deweer et al., 1994). According to the classical interpretation, improved reading time for repeated words is sustained by a repetition priming effect, while procedural learning is demonstrated when this improvement is also observed for new words. Following Masson (1986), the hypothesis tested in the present study was that improved reading of new words could also be due to a repetition priming effect rather than to the acquisition of a mirror-reading skill. Indeed, because the same letters are presented throughout the task, a repetition priming effect for the letters could suffice to explain the improvement in performance. To test this hypothesis, we administered to 30 healthy young and elderly subjects and to 30 AD patients a new mirror-reading task in two phases: an acquisition phase comprising pseudo-words constructed with one part of the alphabet, and a test phase in which both pseudo-words constructed with the same part of the alphabet and pseudo-words constructed with another part of the alphabet were presented. If the new pseudo-words composed with repeated letters were read faster, it would reflect a repetition priming effect; if pseudo-words composed of ‘new’ letters were read faster, it would reflect a procedural learning effect. The results show comparable repetition priming effects in AD patients and in healthy elderly subjects, whereas only healthy subjects showed a procedural learning effect. These results suggest, contrary to previous studies, that the learning of a new perceptual skill may not always be preserved in AD.