Respiratory mechanical impedance measured at a single frequency is of great interest to describe change in airways dimensions with time in young children with asthma. Adequate filtering is mandatory in order to eliminate spurious values at rapid flow transient or during glottis closure. The aim of the study was to test a new filtering procedure. Respiratory impedance was measured at 12 Hz using the head generator technique in 33 asthmatic children aged 3.5-7.5 years. Two separate measurements were obtained at baseline and one 10 min after salbutamol inhalation. Assuming corrupted data would correspond to major distortion in flow oscillation, the tested procedure calculated an index expressing departure of each flow oscillation from an ideal sinusoid (flow shape index filter (FSIF)). Cut-offs ranging from 5 to 20% were compared to the visual identification and 'by hand' elimination of corrupted data (visual filtering). Sensitivity and specificity of FSIF versus visual filtering indicated an optimal range of cut-off values between 10 and 15% (FSIF10 and FSIF15). Compared with visual filtering, respiratory conductance in inspiration (Grs) was marginally but significantly lower with FSIF10 (p < 0.01) but not FSIF15 and had slightly larger (within measurement) coefficient of variation with either FSIF10 or FSIF15 (p < 0.01). Between-measurement repeatability of Grs and Grs change induced by salbutamol were similarly estimated by FSIF10, FSIF15 or visual filtering. It is concluded that FSIF appears as a valid alternative to visual filtering. The 10% cut-off is a good compromise considering sensitivity (72%), amount of rejected data (<20%) and effect on Grs and variability.