The topic of this paper is the detection of errors due to residual faults in software, particularly those with temporary effects. After positioning our approach amongst existing fault tolerance and detection techniques, we propose detection mechanisms for such errors. These mechanisms are designed to detect both data and control flow errors. They can be validated by both formal and fault-injection techniques. In particular, we propose a timed trace technique allowing one to specify the expected software behavior and to instantiate from this specification a generic control-flow checking automaton. The critical algorithms of this automaton are formally proved. To develop these mechanisms, we also propose a design and validation method based on a monitoring specification. Finally, we apply these techniques on two cases of embedded real-time software in order not only to validate them but also to estimate their efficiency and applicability.