Prediction of speed skating performance with a power balance model requires assumptions about the kinetics of energy production, skating efficiency, and skating technique. The purpose of this study was to evaluate these parameters during competitive imitations for the purpose of improving model predictions. Elite speed skaters (n = 8) performed races and submaximal efficiency tests. External power output (Po) was calculated from movement analysis and aerodynamic models and ice friction measurements. Aerobic kinetics was calculated from breath-by-breath oxygen uptake (V̇O2). Aerobic power (Paer) was calculated from measured skating efficiency. Anaerobic power (Pan) kinetics was determined by subtracting P aer from Po. We found gross skating efficiency to be 15.8% (1.8%.). In the 1,500-m event, the kinetics of Pan was characterized by a first-order system as Pan = 88 + 556e-0.0494t (in W, where t is time). The rate constant for the increase in Paer was -0.153 s-1, the time delay was 8.7 s, and the peak Paer was 234 W; Paer was equal to 234[1 - e-0.153(t-8.7)] (in W). Skating position changed with preextension knee angle increasing and trunk angle decreasing throughout the event. We concluded the pattern of P aer to be quite similar to that reported during other competitive imitations, with the exception that the increase in Paer was more rapid. The pattern of Pan does not appear to fit an "all-out" pattern, with near zero values during the last portion of the event, as assumed in our previous model (De Koning JJ, de Groot G, and van Ingen Schenau GJ. J Biomech 25: 573-580, 1992). Skating position changed in ways different from those assumed in our previous model. In addition to allowing improved predictions, the results demonstrate the importance of observations in unique subjects to the process of model construction.