The latent heat flux over snow near Ny-Ålesund, Spitsbergen, was investigated for the 1999 snowmelt season to assess different methods of modelling the flux. Snow evaporation had hitherto been estimated as the residual of plot water balance calculations and was subject to measurement errors: hence a modelling solution was sought to make use of existing data. Precipitation, snow depth and albedo were measured daily. Runoff from the plots was recorded continuously. Wind speed, relative humidity and air temperature were measured at two levels (2 and 10 m) every 10 minutes; wind direction was noted hourly. Three models which simulate latent heat were assessed against evaporation and condensation measured by weighing several snow-filled containers each day. Two employ the bulk profile method (within the SNTHERM and CROCUS snowmelt models), while the third is the aerodynamic profile method (APM). Each follows the measured evaporation until snow-free patches develop after which the APM predicts evaporation whereas the snowmelt models predict condensation. The effect of wind is also noted. A major conclusion of this work is that the complexity of the land surface/atmosphere interactions, particularly when the snow cover breaks up, precludes the use of simple models for determining latent heat.