The existing literature presents conflicting models of how "this" and "that" access different segments of a written discourse, frequently relying on implicit analogies with spoken discourse. On the basis of this literature, we hypothesized that in written discourse, "this" more readily accesses the adjacent/right frontier of a preceding chunk of text, whereas "that" more readily accesses the distant/left. We tested this hypothesis in two eye-tracking experiments, one sentence completion experiment, and one corpus study. Our results showed that both "this" and "that" access the adjacent frontier more easily than the distant. Contrary to existing theories, this accessed the distant frontier more frequently and easily than that. We propose a processing model integrating segmented discourse representation theory's concept of the left/distant leaf with Grosz and Sidner's attentional and intentional model and Garrod and Sandford's focus framework model, suggesting an important role for working memory and emphasizing the different production modes of readers and writers.