When species presence–absence data are analyzed across a range of sites, many communities show a “checkerboard distribution” in which two or more species do not co-occur despite sharing geographic ranges and habitat requirements. It is not clear how the loss of native species might shift the underlying checkerboard structure. I tracked changes in the incidence of 175 vascular plant species across 59 sites in northern Wisconsin that have declined in species density between 1950 and 2000. Based on C-scores and the number of perfect checkerboard distributions, there was significant co-occurrence structure in both time periods. The nature of that structure depended on the particular null model was constructed. When only row values are fixed, plant species showed greater degrees of co-occurrence than the null model predicted, while fixing both rows and columns yielded less co-occurrence than found in the null model. Changes in species density did not strongly influence co-occurrence patterns, reflecting relative stability of co-occurrence patterns over a 50-year interval. This suggests that minor losses of species will not necessarily lead to fundamental changes in co-occurrence structure.