The upper airway (UA) in humans is commonly modeled as a Starling resistor. However, negative effort dependence (NED) observed in some patients with obstructive sleep apnea (OSA) contradicts predictions based on the Starling resistor model in which inspiratory flow is independent of inspiratory driving pressure when flow is limited. In a respiratory bench model consisting of a collapsible tube and an active lung model (ASL5000), inspiratory flow characteristics were investigated in relation to upstream, downstream, and extra-luminal pressures (denoted as P_us, P_ds, and P_out, respectively) by varying inspiratory effort (muscle pressure) from −1 to −20 cmH₂O in the active lung. P_us was provided by a constant airway pressure device and varied from 4 to 20 cmH₂O, and P_out was set at 10 and 15 cmH₂O. Upstream resistance at onset of flow limitation and critical transmural pressure (P_tm) corresponding to opening of the UA were found to be independent of P_us, P_ds, and P_out. With fixed P_tm, when P_ds fell below a specific value (P_ds′), inspiratory peak flow became constant and independent of P_ds. NED plateau flow patterns at mid-inspiration (V̇_n) were produced within the current bench setting when P_ds fell below P_ds′. V̇_n was proportional to P_ds, and the slope (ΔV̇_n/ΔP_ds) increased linearly with P_tm. P_tm and P_ds were the two final independent determinants of inspiratory flow. Our bench model closely mimics a flow-limited human UA, and the findings have implications for OSA treatment and research, especially for bench-testing auto-titrating devices in a more physiological way. NEW & NOTEWORTHY A respiratory model consisting of a collapsible tube was used to mimic a flow-limited human upper airway. Flow-limited breathing patterns including negative effort dependence were produced. Transmural and downstream pressures acting on the tube are the two independent determinants of the resulting inspiratory flow during flow limitation. The findings have implications for obstructive sleep apnea treatment and research, especially for bench-testing auto-titrating devices in a more physiological way.