The refreezing of meltwater is known to be a critical factor when assessing the mass balance state of glaciers in the high Arctic. Several different models of this process are widely used in numerical glacier simulations but, as yet, no attempt has been made to compare these approaches or to assess their relative performance against field measurements. Such models are likely to include implicit calibrations which may prove to be invalid under conditions different from those for which they were constructed. We test a selection of six refreezing algorithms taken from the literature against field measurements of net mass balance and superimposed ice formation made at a high Arctic glacier. Running these models for 1971–2002 using daily meteorological data and forward until 2050 using a range of future scenarios for the climate of Svalbard we are able to make the following conclusions concerning mass balance modeling: (1) The choice of meltwater refreezing model has a significant impact on the calculated net mass balance of a glacier. (2) Three of the models tested could, with suitable tuning, simultaneously reproduce both superimposed ice quantities and net mass balance measurements within their respective bounds of uncertainty, while the other three could not. (3) In situations where the model can be tuned to field measurements indicative of mass balance, future predictions of mass balance are largely independent of the specific meltwater refreezing model used.