The present series of studies examine how jurors and public defenders evaluate different pieces of evidence and integrate them into a coherent conclusion within the context of a criminal case. Previous research has shown that in situations where both sides of the case are compelling, decision-makers nevertheless come to highly confident and polarized decisions, called coherence shifts (Simon, 2004). The present research sought to expand on coherence effects, improve upon the methodology of previous studies, and explore potential moderators of coherence. In Study 1, mock jurors (n = 306) read about a criminal case and evaluated multiple pieces of evidence at various points throughout the case. Results indicated that participants exhibited pronounced coherence shifts (i.e., their evaluations of the various pieces of evidence (a) became more consistent as the case progressed, and (b) were evaluated in line with their initial leanings) using an improved methodology that randomized evidence order and evidence valence. Furthermore, participants’ interim leanings of guilt or innocence biased their subsequent evaluations of ambiguous evidence. The direction and magnitude of participants’ coherence shifts were predicted by their pretrial dispositions towards prosecution and defense. Participants lacked awareness of how their perceptions of the evidence have shifted. Coherence shifts were not, however, moderated by asking mock jurors to justify their decisions, or by asking mock jurors to play devil’s advocate while considering each piece of evidence, underscoring the pervasiveness of this cognitive bias. Study 2 examined whether actual public defenders experience coherence shifts and how those shifts relate to the plea bargaining process; however, no coherence shifts were observed. Study 3 examined whether the timing of the defense’s presentation of their case could reduce coherence effects; results indicated that reading about the defense’s case immediately after the prosecution’s case (c.f. following a delay) marginally (p = .09) reduced coherence effects among jurors who acquitted the defendant, suggesting one potential strategy to mitigate this bias.