Respiratory input impedance (Zrs) was measured from 8 to 256 Hz in 10 healthy subjects by a method that eliminated the shunt impedance of extrathoracic airway walls. It consisted of combining the data obtained with a pressure input at the mouth (standard method, Zst) and with a pressure input around the head (Zhg) Zrs = Zst.(Zp + Zhg)/(Zp + Zst) where Zp is the impedance of the mouthpiece and pneumotachograph. Large quantitative differences were observed between Zrs and Zst, demonstrating that the standard method is unreliable at such frequencies. The real part of Zrs increased from 2.6 +/- 0.8 cmH2O.l-1.s at 8 Hz to a maximum of 38 +/- 19 cmH2O.l-1.s at 158 +/- 49 Hz. The imaginary part exhibited a maximum of 19 +/- 8 cmH2O.l-1.s at 126 +/- 38 Hz, a resonance at 157 +/- 43 Hz, and a minimum of lambda 19 +/- 16 cmH2O.l-1.s at 185 +/- 45 Hz. The data were analyzed with five models featuring alveolar gas compressibility; tissue resistance, inertance, and compliance; and different representations of the airways with lumped and distributed parameters. All except the simplest (lumped frequency-dependent resistance) fitted the data equally well, but none provided reliable estimates of gas compliance. Three models gave a consistent description of the airway in terms of equivalent rigid tubes (cross-sectional area 3.5-3.7 cm2, length 47-51 cm). We conclude that high-frequency input impedance could prove useful in exploring the airways but not the peripheral lung.