Eddy covariance data collected over a horizontal surface on the largest ice body on Kilimanjaro, Tanzania, over 26–29 July 2005 were used to assess the uncertainty of calculating sublimation with a surface energy balance (SEB) model. Data required for input to the SEB model were obtained from an existing automatic weather station. Surface temperatures that were solved iteratively by the SEB model were used to compute emitted longwave radiation, turbulent heat fluxes using the aerodynamic bulk method and the subsurface heat flux. Roughness lengths for momentum and temperature, which were found to be the most important input parameters controlling the magnitude of modelled (bulk method) turbulent heat fluxes, were obtained using eddy covariance data. The roughness length for momentum was estimated to be 1.7×10–3 m, while the length for temperature was one order of magnitude smaller. Modelled sensible and latent heat fluxes (bulk method) compared well to eddy covariance data, with root-mean-square differences between 3.1 and 4.8 Wm–2 for both turbulent heat fluxes. Modelled sublimation accounted for about 90% of observed ablation, confirming that mass loss by melting is much less important than sublimation on the horizontal surfaces of the remaining plateau glaciers on Kilimanjaro.