Assisted ventilation with pressure support (PSV) or proportional assist (PAV) ventilation has the potential to produce periodic breathing (PB) during sleep. We hypothesized that PB will develop when PSV level exceeds the product of spontaneous tidal volume (VT) and elastance (VTsp . E) but that the actual level at which PB will develop [PSV(PB)] will be influenced by the DeltaPCO2 (difference between eupneic PCO2 and CO2 apneic threshold) and by DeltaRR [response of respiratory rate (RR) to PSV]. We also wished to determine the PAV level at which PB develops to assess inherent ventilatory stability in normal subjects. Twelve normal subjects underwent polysomnography while connected to a PSV/PAV ventilator prototype. Level of assist with either mode was increased in small steps (2-5 min each) until PB developed or the subject awakened. End-tidal PCO2, VT, RR, and airway pressure (Paw) were continuously monitored, and the pressure generated by respiratory muscle (Pmus) was calculated. The pressure amplification factor (PAF) at the highest PAV level was calculated from [(DeltaPaw + Pmus)/Pmus], where DeltaPaw is peak Paw - continuous positive airway pressure. PB with central apneas developed in 11 of 12 subjects on PSV. DeltaPCO2 ranged from 1.5 to 5.8 Torr. Changes in RR with PSV were small and bidirectional (+1.1 to -3.5 min-1). With use of stepwise regression, PSV(PB) was significantly correlated with VTsp (P = 0.001), E (P = 0.00009), DeltaPCO2 (P = 0.007), and DeltaRR (P = 0.006). The final regression model was as follows: PSV(PB) = 11.1 VTsp + 0.3E - 0.4 DeltaPCO2 - 0.34 DeltaRR - 3.4 (r = 0.98). PB developed in five subjects on PAV at amplification factors of 1.5-3.4. It failed to occur in seven subjects, despite PAF of up to 7.6. We conclude that 1) a PCO2 apneic threshold exists during sleep at 1.5-5.8 Torr below eupneic PCO2, 2) the development of PB during PSV is entirely predictable during sleep, and 3) the inherent susceptibility to PB varies considerably among normal subjects.