Typical assessment of language in Alzheimer's disease typically relies on experimental tasks such as confrontation naming or fluency tasks. These widely used methods show several advantages such as rapidity of assessment, standardization and good control for a set of important psycholinguistic variables. However some studies suggest that analysis of discourse may be better than naming tasks for assessing everyday word-finding difficulties in elderly adults (Schmitter-Edgecombe et al. (2000). Several psycholinguistic variables are known to affect word retrieval on both normal and pathological populations among which Age of Acquisition (henceforth AoA): words acquired earlier are easier and faster to process than words acquired later (Bates et al., 2001; Hodgson & Ellis, 1998; Lymperopoulo et al., 2006). This variable has so far received little attention in the Alzheimer's population although it has been proposed as an early marker of the disease (Forbes-McKay, et al., 2005) Twenty participants diagnosed with Alzheimer's disease and twenty healthy controls matched for age, sex, level of education and socioeconomic status participated in our study. Spontaneous speech data were elicited by asking participants about biographical details. Any silence exceeding 200 ms was coded as a silent pause. Filled pauses, vocalic lengthenings and hesitations were also measured. These different dysfluencies were used to identify words prone to retrieval difficulty. For each text, a number of problematic words were identified, and a set of the same number of unproblematic words was randomly selected. 49 native French-speaking adults rated AoA on a 7-grade scale for the two word lists. In addition the frequency of each word was computed using the Lexique Database (New et al., 2004). Results reveal that problematic and non-problematic words do not significantly differ in frequency. Similarly no effect of frequency is observed between the two groups of participants. Turning to AoA, we find a group effect: problematic words produced by patients are acquired significantly later than the ones produced by healthy subjects. Similarly, in the Alzheimer's group, problematic words are acquired significantly later than non-problematic words while the difference is not significant in the control group. We discuss our results in the framework of the retrogenesis hypothesis (Reisberg, Franssen, Auer, Akram & Kenowsky, 2002).