Although recent work has investigated the effects of participant—and treatment—related variables on word-retrieval outcomes following semantic-feature based treatments (SFT) in monolingual aphasia, similar research has not been undertaken for bilinguals with aphasia. This retrospective study examined: 1) training outcomes from an SFT protocol for bilinguals with aphasia; 2) patterns of response generalization to untrained items and languages; and 3) the influence of treatment, participant, and item-level factors on treatment efficacy. Participants were 22 Spanish-English bilinguals with aphasia who completed 10 weeks of computerized SFT (Kiran et al., 2013; Peñaloza et al., 2020) in-person or via videoconference (Peñaloza et al., 2021). Treatment progress and response generalization to untrained items were assessed via 3 baseline probes, 10 treatment probes, and 3 post-treatment probes. Probes consisted of 90 items: 15 each of trained items, semantically-related items, and unrelated control items as well as their corresponding translations in the untreated language (e.g. apple-orange-horse and manzana-naranja-caballo). Psycholinguistic properties of treatment stimuli (item-level) and baseline naming severity (participant-level) scores were entered as predictors in a series of logistic mixed-effects models to estimate treatment outcomes. Primary analyses revealed that the probability of a correct response increased more over time for the trained words relative to the controls in both the treated and untreated languages. Secondary analyses demonstrated clear effects of baseline naming severity and psycholinguistic properties of treatment stimuli on naming accuracy over time. This work suggests that a variety of multilevel factors may be used to predict response to treatment in bilingual aphasia.