When operated far from their optimum conditions, pump-turbines may exhibit strong hydrodynamic instabilities, often called rotating stall, which lead to substantial increase of vibration and risk of mechanical failure. In the present study, we have investigated the flow filed in a model of radial pump-turbine with the help of tuft visualization, wall pressure measurement and structure-borne noise monitoring. As the rotation speed is increased, the machine is brought from its optimum operation to runaway with zero torque on the shaft. The runaway operation is characterized by a significant increase of pressure fluctuation at the rotor–stator interaction frequency. As the speed is further increased, the flow exhibits sub-synchronous instability, which rotates at 70% of the rotation frequency. Tuft visualization clearly shows that, as the instability evolves, the flow in a given distributor channel suddenly stalls and switches to reverse pumping mode in periodic way. We have also investigated the monitoring of the rotating stall with the help of vibration signals. A specific signal processing method, based on amplitude demodulation, was developed. The use of 2 accelerometers allows for the identification of the optimum carrier frequency by computing the cyclic coherence of vibration signals. This non-intrusive method is proved to be efficient in detecting the rotating stall instability and the number of stall cells. We strongly believe that it could be implemented in full scale pump-turbines.