For small, microcomputer-based realtime systems, running under a conventional multitasking operating system such as INTEL’s RMX 85/86, a simple recovery strategy is proposed. The strategy does not address the problems of error detection and damage assessment; the provision of tests and alternative paths in case of error is the obligation of the application programmer. Basically, a recovery point is associated to each interaction point among individual tasks. Such interaction points are synchronization (in particular for mutual exclusion) and message communication. The corresponding primitives, signal/wait resp. send/receive, are extended to include the setting of recovery points by means of lower-level primitives SAVE (local and system data) and SET (task control blocks), the latter operating on protected memory banks. The semantics of the extended primitives, in particular their connection to the system RESTART procedure, is described in Petri net notation. In case of any task failure, RESTART resets the whole system to the current recovery line. In view of more recent technological advances, an integration of the proposed approach into a comprehensive strategy of fault tolerance would seem indicated.