On a cue-based retrieval account of sentence processing (Van Dyke & Lewis, 2003; Vasishth & Lewis, 2006), grammatical heads such as verbs provide retrieval cues that are used to distinguish between the target item and competitors in memory. Similarity based interference occurs when items share retrieval cues, which makes it harder to distinguish between them, causing both longer reading times (RTs) and lower question-response accuracy. Since lower accuracy could be the result from either incorrectly retrieving a competitor or simply failing to complete a retrieval (an unstarted or aborted process), it is unclear how RTs are related to question-response accuracy. We investigated this question with two approaches: (i) by using the outcome of multinomial processing trees modeling accuracy in a linear mixed model with RTs as a dependent variable, and (ii) by fitting RTs and accuracy with ACT-R.