Fault-tolerant techniques are often employed to provide predictable performance in the presence of malfunction of embedded real-time systems. Our research addresses the problem of restoring real-time systems to a prior state when a fault generates errors. In this paper, we first define the state restoration problem in real-time systems, explaining when is necessary to restore a real-time system to a prior state. We then make a thorough comparison of two important state restoration schemes, viz. backward error recovery and failure recovery, in an integrated framework. The comparisons give us an in-depth study of both schemes and allow us to characterize the kinds of fault-tolerant problems in real-time applications that each scheme is best suited to deal with. Finally, we propose a novel technique to avoid the domino effect that may arise when backward error recovery takes place.