Four experiments examined 3 types of violations of coherence criteria in risk assessments of a terrorist attack. First, the requirement that extensionally equivalent descriptions be assigned the same probability (i.e., additivity) was violated. Unpacking descriptions of an attack into subtypes led to an increase in assessed risk. Second, additivity was also violated when risk assessments were obtained by subtracting the probability of no attack from 1.0. This refocusing procedure inflated assessed risk. Third, refocusing also increased the proportion of monotonicity violations in assessing risk across increasing or decreasing timeframes. Task structuring that promoted consideration of complementary possibilities increased coherence, suggesting that incoherence is due primarily to errors in applying rather than comprehending the relevant criteria.