The identification of model parameters using measured input-output data of linear systems operating in a closed loop is discussed. A frequency-domain maximum likelihood estimator which takes into account the correlation between the input and output disturbances is presented. Its properties are analyzed and illustrated by simulation examples and real measurements.