A growing body of evidence has shown that repetitive transcranial magnetic stimulation (rTMS) can enhance word-retrieval abilities in chronic aphasia. However, there remains significant variability in the efficacy of combined rTMS and language treatments. This study investigated how semantic and phonological characteristics of baseline word-retrieval impairments may influence the efficacy of rTMS on long-term naming improvements following language treatment in individuals with chronic aphasia. Thirty participants with post-stroke aphasia underwent 10 sessions of 1 Hz rTMS to right pars triangularis followed by a modified constraint-induced language treatment (mCILT). Nineteen participants were randomly assigned to active rTMS and 11 participants were assigned to sham rTMS. All participants completed the Philadelphia Naming Test (PNT) at baseline and at 3 and 6 months post-treatment. We coded PNT errors and fit data to the semantic-phonological (or SP) computational model (Foygel & Dell, 2000) to derive semantic and phonological parameter weights. We ran linear regressions for the proportional improvement in naming, with fixed effects for interactions between rTMS, time, and baseline parameter weights. While there was no immediate effect of rTMS post-treatment, rTMS combined with mCILT improved long-term naming more than language therapy alone. Furthermore, greater baseline semantic and phonological characteristics of word-retrieval abilities were each associated with increased rTMS-induced gains in proportional naming improvements. These patterns were maintained at both 3 and 6 months post-treatment. This study is among the first in a larger sample to demonstrate that individual differences in lexical retrieval contribute to variability in sustained rTMS and aphasia treatment outcomes.