Pink-footed geese (Anser brachyrhynchus) breed in the Arctic, where their populations have doubled since the 1980s. There is concern that nesting geese disturb the fragile tundra and lead to a trophic cascade with strong top-down effects on vegetation and soil processes. A better understanding of the distribution of geese and factors that influence nest site selection is needed to highlight potential problem areas and assess the potential for further population expansion. To help infer the importance of environmental variables on nest site selection, we built generalized additive models using nest observations collected in 2003 and 2004 from the Sassendalen valley, Svalbard, along with a suite of geographical information system explanatory predictors. The fit of the models was very high (explaining over 72% of the deviance), and predictive power to independent samples indicated useful predictions that could discriminate between presences and absence of nests very well (area under the receiver operating characteristic curves exceeded 0.88). Significant predictors of nest site selection included elevation, slope, aspect, percentage of snow cover, percentage of foraging habitat cover, and a spatial autocovariate. Spatial predictions were applied to the broader Nordenskiöldsland region of Svalbard and highlighted the importance of previously unsurveyed locations for nesting.