We present a quantitative comparison of two popular approaches for recovering from CPU errors: Quadruple Modular Redundancy and Backward Error Recovery. Both are used in existing fault-tolerant systems offering basically the same main features and, in particular, the same fault-tolerance services (transparent recovery for hardware faults). We show that the use of performability measures is richer than classical dependability analysis. Given that they take into account not only reliability aspects but also performance metrics, they allow a deeper insight into the behaviour of the considered systems. For instance, they allow the user to identify different mission lengths leading to better adaptation of each type of architecture.