Early detection of dreissenid mussels (Dreissena polymorpha and D. rostriformis bugensis) is crucial to mitigating the economic and environmental impacts of an infestation. Plankton tow sampling is a common method used for early detection of dreissenid mussels, but little is known about the sampling intensity required for a high probability of early detection using the method. We used implicit dynamic occupancy models to estimate plankton tow detection probabilities of dreissenid mussels from a long-term data set containing plankton tow samples collected across central and western United States. We fit models using a) the entire data set, including water bodies with unknown occupancy status in addition to heavily infested water bodies, b) a data subset that included water bodies with paired water temperature data, and c) a data subset that included water bodies with lower dreissenid densities. For the entire data set, we found that estimated detection probabilities varied by water body size and ranged from approximately 0.10 to 0.86. For the water temperature subset, we observed the same pattern between detection probability and water body size as we did for the full data but additionally found that the estimated detection probabilities were much higher when water temperatures were above 12 °C. For the lower dreissenid density subset, we found that the estimated probability of detecting dreissenid mussels with a single aggregated plankton tow sample was near zero. Given these estimates, we conclude that the number of aggregated plankton tow samples taken per water body in the data is far fewer than the number needed to ensure a high probability of detecting dreissenid mussels, especially if they are at low densities. We summarize the analyses with a discussion of plankton tow sampling protocol changes needed to improve estimates of dreissenid detection probabilities.