The paper describes the design, construction and testing of a high-power klystron suitable for an output of 2 MW at 400 Mc/s and originally proposed for operation at 200 microsec pulse length and 50 pulses/sec, though not tested to this mean power.The valve was satisfactorily run up to 3.5 MW at 120kV 60 amp, and detailed contour diagrams of output power against detuning and input power were taken. The theoretical interpretation of these results is given, in particular a comparison with the theory of Warnecke, Guénard and others. A modification was made on the basis of this theory which improved the efficiency of the valve.