This paper presents a model-based fault detection (FD) and isolation scheme for rigid manipulators. A single fault acting on a specific actuator or on a specific sensor of the manipulator is detected (and, if possible, the exact location of the fault), and an estimation of the fault signal is performed. Input-signal estimator and output observers are considered in order to make the FD procedure possible. By using the suboptimal second-order sliding-mode (SOSM) algorithm to design the input laws of the observers, satisfactory stability properties of the observation error are established. The proposed algorithm is verified in simulation and experimentally on a COMAU SMART3-S2 robot manipulator.