Background: Recent studies have indicated that working memory is not a unitary resource and that different types of working memory are used for different types of linguistic processing: syntactic, semantic, and phonological. Phonological working memory was found to support the comprehension of sentences that require re-access to the word-form of a word that appeared earlier in the sentence. Aims: This study explored the relation between phonological working memory and sentence comprehension by testing the comprehension of garden path sentences in individuals with conduction aphasia who have very limited phonological working memory. Our prediction was that if phonological working memory limitation hampers word-form reactivation, only the comprehension of garden paths that require word-form reactivation will be impaired, whereas garden paths that require only structural reanalysis will be better preserved. Methods & Procedures: Five individuals with conduction aphasia and 15 matched controls participated in working memory tests and a garden path comprehension test. The phonological working memory assessment included a battery of 10 tests, which showed that four of the individuals, who had input conduction aphasia, had very limited phonological working memory, and one individual, with output conduction aphasia, had unimpaired working memory. The comprehension study included 60 garden path sentences of three types: structural garden paths, which require only structural reanalysis, lexical garden paths, which require lexical re-access in addition to structural reanalysis, and optional-complement garden paths, which require re-access to the lexical-syntactic frame of the verb in addition to the structural reanalysis. Outcomes & Results: The main result was that the individuals with input conduction aphasia showed different degrees of impairment in different types of garden path sentences. The lexical garden paths were exceptionally hard for them, with a mere 10% correct, and significantly more difficult than the structural garden paths. The individual with output conduction aphasia whose working memory was intact comprehended the lexical garden paths similarly to the normal controls. Conclusions: These findings indicate that phonological working memory impairment only affects the comprehension of sentences that require phonological, word-form re-access. The type of sentence and the type of processing it requires should be taken into account when trying to predict the effect of working memory limitation on a patient's ability to understand sentences. Whereas individuals with input conduction aphasia can understand complex syntactic structures well, they have considerable difficulties understanding sentences that also require re-access to a word-form.