PURPOSE: To determine whether a deviation from linearity occurs in the .VO2-speed relationship, above the lactate threshold (LT) in running; and whether the length of the submaximal exercise bouts alters the magnitude of any deviation. METHODS: Ten endurance-trained runners (N = 3 state level, N = 4 club level, and N = 3 recreational) (mean +/- SE; age 24.4 +/- 2.8 yr, mass 76.1 +/- 2.2 kg, .VO2 59.3 +/- 10.6 mL.kg-1.min-1) completed a .VO2, LT test and 10 x 4-min submaximal constant load exercise bouts. Data were evenly spread above and below LT, which was fitted by a dual linear regression model. RESULTS: There was a significant decrease (51.4%) in slope of the .VO2-speed relationship above LT. The use of 3-min, in comparison with 4-min, submaximal data did not alter the slope of the .VO2-speed regression above LT. There was no significant difference in the .VO2max estimated from the .VO2-speed regression above LT (58.1 +/- 3.3 mL.kg-1.min-1) but a significant difference below LT (63.6 +/- 3.9 mL.kg-1.min-1) to that obtained during the .VO2max test. CONCLUSION: Data from the current study suggest that the use of the linear regression of .VO2-speed data below the LT may potentially overestimate the prediction of .VO2 values above LT.