Airflow and pressure were measured post-operatively in eight mechanically ventilated patients in the routine intensive care unit. Analysis of the input impedance spectra versus frequency suggested that respiratory data cannot be adequately reproduced using the classic two-element R-C model, as the real part of input impedance decreases with frequency. To fit in with this behaviour, we adopted a three-element model with an additional parallel compliance. The three parameters of this model were estimated separately in the frequency and time domains by minimising suitable least-square criterion functions. The results demonstrate a good agreement between the parameter estimates in the frequency and time domains, and show that the three-element model reproduces the input impedance frequency pattern in the range 0.2-8 Hz. Comparison of different linear models in the time domain demonstrated that the precision of parameter estimates and the quality of best fitting sharply increase from the two-element to the three-element model. The addition of a fourth resistive parameter, like in the Mead model, does not lead to appreciable improvement and makes the model almost unidentifiable. The possible contribution of a ventilator-patient circuit of the upper airway shunting and of the peripheral airway obstruction are also discussed.