In this dissertation, I examined children’s ability to generate two types of pragmatic inferences during online language comprehension. Using referential communication tasks combined with eye-tracking technology, I assessed children’s online pragmatic inferencing in two situations: 1) when inferences are assumed to be context-independent (Chapter 2); and 2) when inferences are closely related to particular aspects of a communicative context (Chapter 3). In Chapter 2, 4- and 5-year-old children completed a referential communication task with either: (1) a conventional speaker who used language in a conventional way; or (2) an unconventional speaker who was introduced as saying things in a weird way and who used language in an infelicitous and non-optimal way. On critical trials, children were asked to look at an object (e.g., “Look at the big duck”) while presented with displays involving a target, a competitor, and either a contrast object (Contrast trials) or another distractor object (No contrast trials). When interacting with a conventional speaker, both 4- and 5-year-old children rapidly generated contrastive inferences upon hearing prenominal size adjectives. That is, children made anticipatory fixations toward the target object when a contrast object existed on the display (e.g., small duck). When there was no contrast object, children showed a tendency to look at the competitor object (i.e., the other big object) during the unfolding adjective. When interacting with the unconventional speaker, children showed a delayed preference to look at the target over the competitor during contrast trials. In Chapter 3, I examined children’s online pragmatic inferencing in a situation that involved knowledge discrepancies between children and a speaker. In Experiment 1, 5-year-old children either possessed privileged knowledge or shared knowledge about object identity with a speaker. Children’s tendency to consider a target object (e.g., candle) over a competitor object (e.g., a candle that looks like an apple) was assessed in a referential communication task. In Experiment 2, children’s performance in a condition where the knowledge mismatch was associated with object identity was compared to performance in a condition where the knowledge mismatch related to the physical presence of objects. Results demonstrated children’s egocentric bias to consider their privileged knowledge about the actual identities of deceptive objects, rather than taking a speaker’s lack of knowledge into account. However, when the knowledge mismatch was created by the awareness of the physical presence of objects, children rapidly considered a speaker’s knowledge to guide their interpretation. Together, these findings demonstrated that preschool children make distinct kinds of inferences to understand a speaker’s intended meaning in online comprehension. The efficiency of their online inferencing, however, depends on the specific type of mental representation required in a context. Chapter 4 summarizes and explores the current findings in greater detail.