A method for providing fault tolerance in a client/server computer system. A client is initially connected to a primary server. The primary server normally handles all of the client's transactions. However, a secondary connection is designated to a backup server, whereby if the primary server ever becomes disabled, the client is automatically switched over to the backup server. In-flight transaction information corresponding to transactions currently being processed by the primary server is stored and regularly updated by the client. The client regularly polls the primary server to check whether the primary server is properly functioning. If the primary server becomes disabled, all in-flight transactions pending on the disabled server are rolled back and the client resubmits in-flight transaction information to the backup server so that it can complete any transactions which were in progress on the primary server at the time of the failure. Thus, server failures are transparent to the user.