This study investigated the stability of the measurement of respiratory variables during rest, walking and running using the K4b(2) portable metabolic analyser in ten active males (age 31 ± 11 years; VO₂ peak 42.1 ± 2.6 ml · min(-1) · kg(-1)). Following a 10 min rest, participants completed three discontinuous incremental exercise tests on a treadmill while walking (4, 5 and 6 km · h(-1)) and running (8, 10, 12, 14 and 16 km · h(-1), or until volitional exhaustion). Participants completed 3 min of exercise at each speed, followed by a 3 min recovery after each stage above 10 km · h(-1). The respiratory variables were measured using either a laboratory-based metabolic cart as a reference method (Oxycon Pro, OP), a K4b(2) calibrated immediately before the test (K4b(2)), or a K4b(2) calibrated 1 h before the test (K4b(2)DEL). Compared to the OP, carbon dioxide production (VCO₂) and V(E) were not different when measured by K4b(2) or K4b(2)DEL. There was no difference in VO₂ between OP and K4b(2) tests (P = 0.19, mean difference = 38 ml · min(-1) and limits of agreement (LOA) = 208 to -285) although K4b(2)DEL overestimated VO₂ (P = 0.05, mean difference = 84 ml · min(-1) and LOA = 302 to -469). These data suggest that a drift in measurement accuracy appears to cause the K4b(2) to overestimate VO₂ in tests lasting longer than 1 h.