A class-bridge rule is the rule whose antecedent and consequent belong to different conceptual classes. This kind of rules stands for the correlation between conceptual classes. The study on class-bridge rules can benefit many domains such as criminal detection, chemical synthesis, biological grafting, etc. Class-bridge rules are different from other association rules as follows: (1) class-bridge rules can be generated from infrequent itemsets; (2) measurements of class-bridge rules include the distance between conceptual classes and the relation between the antecedents/consequents and their affiliated conceptual classes. This paper proposes an algorithm based on rough sets to mine and evaluate class-bridge rules. The experiment demonstrates its effectiveness.