Although poor comprehenders struggle with word-level semantics, little is known about how they negotiate "higher-level" combinatorial semantic processing. This study investigated the relationship between reading comprehension ability and the processing of expressions requiring enriched semantic composition. Participants completed a battery of reading-related cognitive assessments, and we used a regression-based approach to calculate their comprehension ability relative to component skills. They also completed an eye movement study, reading sentences requiring enriched semantic processing, viz. complement coercion (e.g., Traxler, Pickering, & McElree, 2002). Coercing verbs have semantic selectional requirements, i.e., their complements must denote an event (e.g., started the race); when coercing verbs combine with non-event complement NPs (e.g., started the book), an event-interpretation must be selected (e.g., reading), and processing difficulty results. Our study replicates and extends Frisson & McElree (2008), which tested whether this cost reflected difficulty selecting an appropriate semantic interpretation, and reported that the number of alternative senses did not modulate processing effort – but did not account for their participants' reading ability. We predicted modulating effects of reading ability because poor readers are known to be sensitive to interference from competing, context-inappropriate semantic senses. Consistent with this, we found that poor readers processed coercing expressions more slowly than skilled readers. Further, skilled readers processed coerced expressions with few competing event-senses without apparent cost. This study shows, for the first time, that poor reader's online semantic processing difficulties extend beyond the level of lexical semantics, and that the coercion cost is at least partly attributable to semantic sense selection.