The aim of this work was to assess the accuracy of thermistors/thermocouples as devices for detecting hypopnoeas in sleep studies. Conventional thermistor/thermocouples were studied with a respiratory model allowing the simulation of inspiratory (22 degrees C) and expiratory (37 degrees C) flows. The thermistor signal (V'th) was compared with a pneumotachograph (V'th): 1) for sinusoidal and square-wave airflows (+/-0.05 to +/-0.8 L.s(-1), 10-20 breaths.min(-1) (bpm)); 2) when changing the distance from the thermistor to the nose (0-20 mm); and 3) when doubling the section of the nostrils. The thermistor was strongly nonlinear and flow reductions (hypopnoeas) were underestimated: a 50% reduction in V' (+/-0.5 L.s(-1), 15 bpm, sinusoidal) resulted in only an 18% reduction in V'th. V'th depended considerably on the airflow pattern: for V'=+/-0.5 L.s(-1), V'th increased by 100% from sinusoidal (20 bpm) to square-wave (10 bpm). For V'=+/-0.5 L.s(-1), 15 bpm, sinusoidal flow, V'th increased by 79% when the distance thermistor-nose varied from 20-0 mm, and V'th decreased by 37% when doubling the nose section. We concluded that thermistor/thermocouples are inaccurate flow-measuring devices when used at the airflow conditions typical of sleep studies. Their use for quantifying hypopnoeas may lead to considerable underdetection of these respiratory events.