Observational data allowing the validation of jökulhlaup models are sparse. We were able to inject dye tracer directly into the drainage channel of a glacial lake during the onset of its outburst. This made it possible to test an established jökulhlaup model, not only against discharge measurements, but for the first time also against water flow speeds inferred from measurements. We drive the jökulhlaup model, based on the Spring–Hutter equations, with measured subglacial water pressure, lake water temperature and lake level. The model is fitted to the measured lake discharge and inferred flow speeds using the initial channel size, the channel roughness and sinuosity. Our calculations show that an ingenuous application of the model, fitting it to the lake discharge only, overestimates water flow speeds. For the second day of the outburst, this can be remedied by fitting the model to the inferred flow speeds as well, requiring that either the heat transfer or the sinuosity of the channel be increased. However, the low inferred flow speeds on the first day of the outburst cannot be fitted with any parameter combination, showing that, initially, the water does not flow through an R channel. Hence, the early stages of this jökulhlaup cannot be simulated by an R-channel model.