This paper presents an analysis of a model reference adaptive system (MRAS) observer for the sensorless control of a standalone doubly fed induction generator (DFIG). The analysis allows the formal design of the MRAS observer of given dynamics and further allows the prediction of rotor position estimation errors under parameter mismatch. The MRAS observer analysis is experimentally implemented for the vector control of a standalone DFIG feeding a load at constant voltage and frequency. Experimental results, including speed catching of an already spinning machine, are presented and extensively discussed. Although the method is validated for a standalone generator, the proposed MRAS observer can be extended to other applications of the doubly fed induction machine.