In order for palaeontological data to be informative to ecologists seeking to understand the causes of today's diversity patterns, palaeontologists must demonstrate that actual biodiversity patterns are preserved in our reconstructions of past ecosystems. During the Late Cretaceous, North America was divided into two landmasses, Laramidia and Appalachia. Previous work has suggested strong faunal provinciality on Laramidia at this time, but these arguments are almost entirely qualitative. We quantitatively investigated faunal provinciality in ceratopsid and hadrosaurid dinosaurs using a biogeographic network approach and investigated sampling biases by examining correlations between dinosaur occurrences and collections. We carried out a model-fitting approach using generalized least-squares regression to investigate the sources of sampling bias we identified. We find that while the raw data strongly support faunal provinciality, this result is driven by sampling bias. The data quality of ceratopsids and hadrosaurids is currently too poor to enable fair tests of provincialism, even in this intensively sampled region, which probably represents the best-known Late Cretaceous terrestrial ecosystem on Earth. To accurately reconstruct biodiversity patterns in deep time, future work should focus on smaller scale, higher resolution case studies in which the effects of sampling bias can be better controlled.