A transaction model provides a framework for concurrent processing of retrieval and update operations in a database. Considerable research effort has focused on various techniques and protocols to ensure the ACID properties of transactions in conventional databases. However, the adoption of these techniques and protocols to temporal databases is not trivial. In particular, a refined locking mechanism based on temporal characteristics can provide better concurrency among transactions in temporal databases than a conventional locking mechanism. Accordingly, this paper presents a set of modifications and fine tuning of traditional concepts in transaction management, to enable a better performance of temporal databases. We also suggest a scheme for implementing a transaction protocol for temporal databases on top of a relational database. The contribution of the paper is in identifying the unique properties of transaction management in temporal databases and the use of these properties to provide a refined locking mechanism to enhance the concurrency of such databases. In particular, we show that the classic 2PL mechanism cannot ensure serializability in temporal databases. Instead, we suggest an alternative method to ensure serializability and reduce redundant abort operations, which is based on a temporal serializability graph.