There is growing interest in the use of gridded population models which potentially offer advantages of stability through time and ease of integration with nonpopulation data sources. This paper assesses the accuracy of models of the type introduced by Martin in 1989. Population counts for census output areas (OAs) are reallocated to a 100�m grid and then compared with true 100�m cell population counts uniquely available from the 2001 Northern Ireland Census. This analysis is novel, being the first large-scale assessment of gridded population models against true gridded population counts. We find evidence that kernel width and cell size are more important than the distance-decay parameter; that local mass preservation approaches are more appropriate in urban areas; but that the spatial scale of input data is more important than model parameters. It is suggested that more attention needs to be given to the varying spatial structures of population between places and that incorporating this information through geostatistical approaches could yield further insights.