A think-aloud method was used to examine the content of information available to working memory during narrative comprehension in a CHI population. Twenty severe CHI participants (>1 year post-injury) and 20 controls talked aloud after they read each sentence of story narratives. Trabasso and Magliano's (1996a) verbal protocol analysis was then used to code for the production of inferential and non-inferential clauses and the memory operations that supported inferential clause production. We found that CHI and control groups produced a comparable number of clauses, that inferences dominated narrative comprehension, and that both groups produced more explanatory inferences than predictive or associative inferences. Despite these qualitative similarities, the CHI group demonstrated poorer comprehension, generated proportionately fewer inferences, relied less on retrieval as a memory source for explanatory inferences, and produced more non-inferential clauses and associative inferences.