Research on emotion inferences has shown that readers include a representation of the main character's emotional state in their mental representations of the text. We examined the specificity of emotion representations as a function of the emotion content of short narratives, in terms of the quantity and quality of emotion components included in the narratives, based on the GRID instrument (Fontaine, Scherer, Roesch, & Ellsworth, 2007; Fontaine, Scherer, & Soriano, 2013; Scherer, 2005). In a self-paced reading task, target sentences that only moderately matched the emotional context were read faster than target sentences that strongly matched the emotional context of the narratives. In a "makes sense" judgment task, we showed that this result was not driven by a mapping difficulty, and in a memory task, we provided some evidence that these effects reflected integration processes. We suggest that readers can integrate specific emotions into their mental representations, but only if provided with the appropriate emotional contextual support.