Modeling evapotranspiration (ET) in Arctic coastal plain ecosystems is challenging owing to the unique conditions present in this environment, including permafrost, nonvascular vegetation, and a large standing dead vegetation component. In this study the ecosystem process model, BIOME-BGC, was adapted to represent these unique conditions in Arctic ecosystems by including a new water storage and evaporation routine that accounts for nonvascular vegetation and the effects of permafrost, adding ground heat flux as an input, and representing ground shading by dead vegetation. The new Arctic version and the original BIOME-BGC models are compared to observed ET from two eddy flux towers in Barrow, Alaska over four summer seasons (1999--2002). The two towers are located less than 1 km apart, yet represent contrasting moisture conditions. One is located in a drained thaw lake, marsh area, while the other is located in a drier, upland area characterized by mesic tundra. Results indicate that the original BIOME-BGC model substantially underestimated ET, while the Arctic version slightly overestimated ET at both sites. The new Arctic model version worked particularly well at the wet tower because the model was able to capture energy limitations better than water limitations. Errors in the simulation of snowmelt date led to errors in the ET estimates at both sites. Finally, the substantial differences in soil moisture led to substantially different ET rates between the sites that were difficult to simulate and indicates that soil moisture heterogeneity is a strong controller on ET in these ecosystems.