Background. Word learning ability has been associated with treatment outcomes in people with aphasia (PWA) (Dignam et al., 2016). Our previous work has provided evidence for spared learning abilities in some participants with chronic aphasia. Here, we examined new word learning under word-referent ambiguity in two people at their early stages of post-stroke recovery. Methods. Two PWA (APH1, APH2, both with severe aphasia) and 12 healthy adults (mean age=65.9 range= 60-70) completed a computerized word learning task (Peñaloza et al., 2016) modified to include both auditory and orthographic input and a final oral naming task. Participants were to learn 6 novel word-referent associations (6 trisyllabic pseudowords and 6 unfamiliar object drawings, Laine & Salmelin, 2010). The learning phase included 7 learning blocks (30 trials per block). In each trial, two novel objects (target and foil) appeared on the screen together with the spoken and written pseudoword corresponding to the target object. Participants were to decide whether the word was the label for the object on the left or on the right by pressing mouse button. Visual feedback on correctness was provided after the response. The recognition of the word-referent associations was evaluated immediately after training (test) and at 1-week follow-up, using 30 trials with no feedback. After both two tests an oral naming task was administered. All scores were contrasted against chance level using the exact binomial test. Results. The control group performed significantly above chance from block 5 onwards (p≤ .013). None of the controls could name all the targets when tested immediately after or 1 week post learning. APH1 did not show significant learning in block 7 or in the immediate test, but did so at 1-week follow-up (27/30, p< .001). APH2 demonstrated significant learning with 21/30 correct at block 7 (p= .013) and 20/30 in the immediate test (p= .028), but not anymore at follow-up. The PWA could not name any of the targets after learning. Discussion. Two participants with severe non-chronic aphasia showed significant learning as measured by recognition memory either during the last learning block (block 7) where feedback was still provided (APH2), the immediate test block (APH2) or at 1 week post-learning (APH1). Both participants’ accuracy declined when feedback was removed. For APH2 the decline continued to 1-week follow-up whereas APH1 showed a steep rise to 90% accuracy. The learning pattern of APH1 suggests possible interference caused by the referential ambiguity of the learning phase, which may have cleared up during maintenance to enable the consolidation of new word representations in memory. APH2’s performance decline in the immediate test and follow-up suggests learning that relied heavily on feedback. Both participants had more severe aphasia than those of the previous study (Peñaloza et al., 2016), showing that some preservation of receptive word learning is possible in severe language disorder in the early stages of post-stroke recovery, even under word-referent ambiguity.