The present research explores the possibility of several types of memory representations for sentences. A recognition experiment was performed to test participants’ memory for different types of sentences. Two groups of study sentences were constructed. One group was based around associating pairs of words together in the study sentences that would also appear together in distractor test sentences. The other group’s study sentences maintained the same number of words in common with the distractor test sentences, but with fewer associations between them. It was found that participants were significantly more likely to false alarm to distractor test sentences when they read sentences containing multiple inter-word associations than when they read sentences containing fewer of those associations. A second recognition experiment was performed to further manipulate the ratio of associations in the study sentences and to get a baseline measure of false alarm rates for sentences containing no associations.