The current study investigates how pronominal ambiguity is resolved in real-time, focusing on the role of referent bias and task context. In two self-paced reading experiments, we tested whether ambiguity leads to processing benefits or costs modulated by the presence of a biased referent and the task manipulations. Experiment 1 showed that the ambiguity advantage emerges only when a biased referent is not selected, supporting reanalysis-based accounts such as the unrestricted race model (Van Gompel et al., 2000, 2001, 2005). Experiment 2, however, revealed a delayed ambiguity penalty, suggesting task-induced shifts in processing strategy that better fit a delayed interpretation account. These findings highlight that pronominal ambiguity resolution may involve two processing mechanisms shaped by the parser's evaluation space and the timing of selection.