Sixteen young, healthy males each performed five to seven randomly assigned, exhaustive exercise bouts on a cycle ergometer, with each bout on a separate day and at a different power, to compare estimates of critical power (PC) and anaerobic work capacity (W') among five different models: t = W'/(Pmax-PC) (two-parameter nonlinear); t = (W'/P-PC))-(W'/(Pmax-PC)) (three-parameter nonlinear); P.t = W' + (PC.t) (linear (P.t)); P = (W'/t) + PC (linear (P)); P = PC + (Pmax-PC)exp(-t/tau) (exponential). The data fit each of the models well (mean R2 = 0.96 through 1.00 for each model). However, significant differences among models were observed for both PC (mean +/- standard deviation (SD) for each model was 195 +/- 29 W through 242 +/- 21 W) and W' (18 +/- 5 kJ through 58 +/- 19 kJ). PC estimates among models were significantly correlated (r = 0.78 through 0.99). For W', between-model correlations ranged from 0.25 to 0.95. For a group of six subjects, the ventilatory threshold for long-term exercise (LTE Tvent; 189 +/- 34 W) was significantly lower than PC for all models except the three-parameter nonlinear (PC = 197 +/- 30 W); PC for each model was, however, positively correlated with LTE Tvent (r = 0.69 through 0.91).(ABSTRACT TRUNCATED AT 250 WORDS)