Sleep unmasks a highly sensitive hypocapnia-induced apnoeic threshold, whereby apnoea is initiated by small transient reductions in arterial CO2 pressure (PaCO2) below eupnoea and respiratory rhythm is not restored until PaCO2 has risen significantly above eupnoeic levels. We propose that the ‘CO2 reserve’ (i.e. the difference in PaCO2 between eupnoea and the apnoeic threshold (AT)), when combined with ‘plant gain’ (or the ventilatory increase required for a given reduction in PaCO2) and ‘controller gain’ (ventilatory responsiveness to CO2 above eupnoea) are the key determinants of breathing instability in sleep. The CO2 reserve varies inversely with both plant gain and the slope of the ventilatory response to reduced CO2 below eupnoea; it is highly labile in non-random eye movement (NREM) sleep. With many types of increases or decreases in background ventilatory drive and PaCO2, the slope of the ventilatory response to reduced PaCO2 below eupnoea remains unchanged from control. Thus, the CO2 reserve varies inversely with plant gain, i.e. it is widened with hyperventilation and narrowed with hypoventilation, regardless of the stimulus and whether it acts primarily at the peripheral or central chemoreceptors. However, there are notable exceptions, such as hypoxia, heart failure, or increased pulmonary vascular pressures, which all increase the slope of the CO2 response below eupnoea and narrow the CO2 reserve despite an accompanying hyperventilation and reduced plant gain. Finally, we review growing evidence that chemoreceptor-induced instability in respiratory motor output during sleep contributes significantly to the major clinical problem of cyclical obstructive sleep apnoea.