There are two competing theoretical frameworks with which cognitive sciences examines how people reason. These frameworks are broadly categorized into logic and probability. This paper reports two applied experiments to test which framework explains better how people reason about evidence in criminal cases. Logical frameworks predict that people derive conclusions from the presented evidence to endorse an absolute value of certainty such as ‘guilty’ or ‘not guilty’ (e.g., Johnson-Laird, 1999). But probabilistic frameworks predict that people derive conclusions from the presented evidence in order that they may use knowledge of prior instances to endorse a conclusion of guilt which varies in certainty (e.g., Tenenbaum, Griffiths, & Kemp, 2006). Experiment 1 showed that reasoning about evidence of prior instances, such as disclosed prior convictions, affected participants’ underlying ratings of guilt. Participants’ guilt ratings increased in certainty according to the number of disclosed prior convictions. Experiment 2 showed that participants’ reasoning about evidence of prior convictions and some forensic evidence tended to lead participants to endorse biased ‘guilty’ verdicts when rationally the evidence does not prove guilt. Both results are predicted by probabilistic frameworks. The paper considers the implications for logical and probabilistic frameworks for reasoning in the real world.