SystemC is a popular language used in modeling systemonchip implementations. To support this task at a high level of abstraction, transaction-level modeling (TLM) libraries have been recently developped. While TLM libraries are useful, it is difficult to capture the reactive nature of certain transactions with the constructs currently available in the SystemC and TLM libraries. In this paper, we propose an approach to specify and verify reactive transactions in SystemC designs. Reactive transactions are different from TLM transactions in the sense that a transaction can be killed or reset. Our approach consists of: (1) a language to describe reactive transactions that can be translated to verification monitors, (2) an architectural pattern to implement reactive transactions, and (3) the verification support to verify that the design does not deadlock, allows only legal behaviors and is always responsive. We illustrate our approach through an example of a transactional memory system where a transaction can be killed or reset before its completion. We identify the architectural patterns for reactive transactions. Our results demonstrate the feasibility of our approach as well as support for a comprehensive verification using RuleBase/NuSMV tools.