Actual and reference evaporation from a wet grassland in Southeast England was studied over the spring and summer of 1999 (March to September) through changes in surface wetness. The Penman-Monteith (using resistance values for reference grass surface), equilibrium evaporation and Priestley-Taylor models were compared with output from the Bowen ratio energy balance (BREB) method. On field visits, inundation of the grazing marsh was mapped and surface soil moisture monitored in a regular grid using a capacitance probe. During the study period, the extent of flooding fell from approximately 10% to 0% and the surface soil moisture declined from over 38% to 15%. Daily averaged Bowen ratios displayed large variation but were below unity, indicating that latent energy flux was the dominant energy sink. The Penman-Monteith and equilibrium evaporation models underestimated during periods of surface inundation and overestimated when no surface water was present. Computed values of the Priestley-Taylor parameter α showed daily variability, but α was predictable with surface wetness such an average value of α = 1.25 characterized periods of inundation, and an average values of α = 0.80 represented periods of no surface water. The performance of the models in computing actual evaporation was compared with the BREB using the root-mean-square error and index of agreement. The optimal model was the Priestley-Taylor model. © 2004 John Wiley and Sons, Ltd.