We examined temporal stability of minnow species co-occurrence using four sets of collections made by seining in streams of the central United States. These collections were made at spatial scales ranging from within a single watershed to across a six-state region of the lower Great Plains, and over time periods ranging from 17 to 75 years. A Jaccard's index of co-occurrence based on presence-absence was calculated for all possible species pairs across all sites for each set of collections. Stability of co-occurrence patterns was assessed using Mantel tests to compare co-occurrence matrices for sequential collections. Within a single watershed (Piney Creek, Izard County, Arkansas) patterns of co-occurrence at 12 sites were correlated highly across sequential collections (r = 0.618 to 0.871, n = 9), suggesting that species associations within the watershed are stable and predictable. Within the Red River basin (Red and Washita rivers), correlations of co-occurrence patterns at six sites were variable and some correlations were extremely low across sequential collections (r = −0.153 to 0.870, n = 17). At the largest geographic scale, species co-occurrence patterns were highly correlated for collections made 17 years apart at 61 sites throughout the lower Great Plains (r = 0.828). Co-occurrence patterns also were compared between collections made by A. I. Ortenburger at 48 sites in the Red and Canadian river drainages of Oklahoma in 1925–1927 and collections made by W. J. Matthews at 57 sites in the same drainages in 1975–1999. Patterns of species co-occurrence remained significantly correlated (r = 0.563) over a timespan of greater than half a century, in spite of dramatic changes in the streams, such as construction of large dams.