Plasma lactate and catecholamine thresholds were calculated using three different variations of linear regression, an algorithmic linear regression method, a log-log transformation method and a semi-log method. A group of 18 male sports science students undertook an incremental test to exhaustion on a cycle ergometer. A 5-ml blood sample was drawn at rest, after 4 min of exercise and every 2 min thereafter until the cessation of the test. Lactate, adrenaline and noradrenaline concentrations were measured. Lactate threshold (Th1a), adrenaline threshold (ThA) and noradrenaline threshold (ThNA) were calculated using each of the three methods. The best fits of the methods were examined by comparing their standard error of estimates (SEEs). The algorithmic method demonstrated a higher SEE than the other two methods, but only for Th1a and ThNA. The power output for which each method calculated the thresholds demonstrated a main effect for method. Tukey post hoc tests showed that the algorithmic method produced significantly higher outputs than the other two methods, which did not differ significantly from one another. Comparison of these power outputs showed that Th1a and ThA differed significantly, regardless of method, there were no other significant differences. Plasma concentrations of lactate, adrenaline and noradrenaline showed that the values of Th1a and ThNA calculated by the algorithmic method were significantly higher than those calculated using the other two methods, which did not differ significantly from one another. The only significant difference for ThA was between the algorithmic and semi-log methods. Correlations between the power outputs at which each method calculated the thresholds varied greatly between methods, and were at best only moderate (r = 0.63). It was concluded that the algorithmic method was less powerful than the other two methods, and that Th1a, ThA and ThNA are not highly correlated.