Semantically-oriented naming treatments such as Semantic Feature Analysis (Boyle & Coelho, 1995) can improve both word retrieval and broader communicative function among people with aphasia (PWA). However, there is considerable variability in how well individual PWA respond to SFA treatment (Boyle, 2010; Oh et al., 2016). This talk presents three complementary lines of evidence aimed at understanding and characterizing this variability. First, results from a novel meta-analysis of SFA treatment studies (Quique et al., 2017) provide evidence for a person-level predictor of treatment response, as well as preliminary evidence regarding the dose-response relationship for SFA: how much benefit may be expected from varying amounts of SFA treatment? Second, results from a large-scale on-going group study of SFA response (Gravier et al., 2018) provide evidence for a practice-related predictor of SFA response: the number of client-generated features during treatment appears to be predictive of gains for both treated and untreated related stimuli. Third, parallel results from the same group study (Dickey et al., in prep) indicate that pre-treatment semantic processing ability is predictive of improvement on both treated and untreated words. Together with the findings regarding practice-related predictors, these results suggest that SFA has its positive effects through facilitation of lexical-semantic aspects of word retrieval processes (Foygel & Dell, 2000). Understanding the variability in SFA treatment response can therefore shed novel light on the mechanisms behind it.