Word stem completion and word identification were used in two repetition priming experiments to evaluate the implicit memory performance of Alzheimer's disease (AD) patients. This issue was also approached using various meta-analyses combining and contrasting previously reported data. While the experimental results suggested that AD patients present preserved repetition priming in both tasks, the meta-analytic approach showed an impairment in stem completion in comparison to word identification. Converging evidence cautiously suggested to accept the results of the meta-analysis. The above dissociation has been interpreted as showing differences in the specific contribution of data- and conceptually-driven processes in the two implicit tasks. A further meta-analysis on the effect of reduced perceptual availability of the study material on the same two tasks indicated that this variable affected repetition priming in word identification more heavily than in stem completion. The impact of such a dissociation on theories of implicit memory is discussed.