Various methods of assessing infant chemoreceptor responses have been reported in the literature. However, equipment dead space, trigeminal stimulation and inherent respiratory variability may have affected the results. A method is presented which attempts to reduce the effect of these factors and thereby isolate the chemoreceptor response. Inspiratory gas was delivered into a lightweight face mask with a pliable rim, minimal dead space and a connected pneumotachograph. Ventilatory data were computed breath by breath. Computer-controlled electromagnetic valves allowed instantaneous switching between air and different gas mixtures, repeated in a randomized sequence. In 18 healthy term neonates, the mask increased ventilation by 12% (95% confidence interval 6-18%), measured by calibrated strain-gauge bands. The effect on respiratory frequency and tidal volume differed significantly between sleep states. Neonates were challenged with short-lasting hyperoxia, mild hypoxia, rebreathing and mild hypercapnia. Coherent averaging of several ventilatory responses from each sleep state reduced the variability while maintaining a high time-resolution.