The purpose of the present study was to establish the most appropriate allometric model to predict mean skiing speed during a double-poling roller skiing time-trial using scaling of upper-body power output. Forty-five Swedish junior cross-country skiers (27 men and 18 women) of national and international standard were examined. The skiers, who had a body mass (m) of 69.3 ± 8.0 kg (mean ± s), completed a 120-s double-poling test on a ski ergometer to determine their mean upper-body power output (W). Performance data were subsequently obtained from a 2-km time-trial, using the double-poling technique, to establish mean roller skiing speed. A proportional allometric model was used to predict skiing speed. The optimal model was found to be: Skiing speed = 1.057 · W (0.556) · m (-0.315), which explained 58.8% of the variance in mean skiing speed (P < 0.001). The 95% confidence intervals for the scaling factors ranged from 0.391 to 0.721 for W and from -0.626 to -0.004 for m. The results in this study suggest that allometric scaling of upper-body power output is preferable for the prediction of performance of junior cross-country skiers rather than absolute expression or simple ratio-standard scaling of upper-body power output.