This research takes the stress deafness hypothesis as a starting point (e.g. Dupoux et al., 2008), and, more specifically, the fact that French speakers present difficulties in perceiving lexical stress in a free-stress language. In this framework, we aim at determining whether a prosodic training could improve the ability of French speakers to identify the stressed syllable in Spanish words. Three groups of participants took part in this experiment. The Native group was composed of 16 speakers of Spanish; the Non-Native and Control group were formed of, respectively, 22 and 14 speakers of French with no knowledge of Spanish. The experiment involved three phases. During the pre-test phase, participants performed an identification task in which they heard trisyllabic Spanish words (e.g. número, numero, numeró) and had to indicate the stressed syllable. During the second phase, participants learned triplets of Spanish pseudowords based on accentual contrasts (e.g. lúguido, luguido, luguidó) by associating these pseudowords with visual shapes (e.g. Dufour et al., 2010). In the post-test phase, the same procedure as the pre-test was used to assess changes in the identification of stress. The Control participants, instead of receiving prosodic training between the pre- and post-test phases, had to click on visual shapes appearing on a computer screen, with no oral input. We analyzed participants' responses (correct/incorrect) as well as reaction times, using mixed-effects models (Baayen et al., 2008) in order to examine the influence of training not only on the ability to perceive lexical stress, but also on the time necessary to process the accentual information. Moreover, given that not all participants were equally able to learn the accentual contrasts during the training session, we also took this variability into account in the analysis. Results are discussed in the framework of existing research on stress deafness and models of L2 acquisition.