Objectives: Sentence comprehension requires linguistic information to be maintained in a verbal working memory store whilst thematic role assignment is being undertaken. A current divisive issue in psycholinguistic research centres around the precise nature of this memory resource. Two distinct models have been proposed: 1) a general memory resource exists to support the processing of both sentences and non-syntactic information, and 2) a sentence-specific working memory pool facilitates only the processing of syntax. The present study aims to provide support for the former model by investigating patients with post-stroke language disorders (aphasia), who are unable to comprehend sentences involving complex syntactic structures. Method: An integrated sentence-picture matching and memory load dual-task was presented to 55 healthy control participants, 4 aphasics and 6 non-aphasic stroke patients. Sentences differed in complexity, from simple actives to centre-embedded subject-relatives (SRs). Memory load nouns were either animate or inanimate, to produce matched or unmatched dual-task conditions, to examine similarity-based interference effects in comprehension accuracy. Results: Aphasics comprehended complex sentences less proficiently than controls and non-aphasics, with SRs eliciting only chance-level accuracy (47%). Their memory load recall was more accurate on complex sentence trials, tentatively supporting a resource-reduction explanation for their inability to process complex sentences. An operational effect of task matching was found in young controls, with a facilitatory effect of matching on the comprehension of object relatives (ORs). Conclusion: The disparate effect of high verbal working memory load on comprehension and recall in aphasics, combined with a similarity-based interference effect in complex relative sentences for young controls, is most effectively explained by the general verbal working memory resource model.