Hydrochorous dispersal through agricultural channels plays a role in structuring plant communities across agricultural landscapes. To date, research on seed retention in vegetated areas has mainly focused on vegetation types with simple architecture (often cylinders), which consequently do not represent real vegetation features. Here, we test the hypothesis that vegetation cover estimated at the water surface best explains floating seed retention in open channels. We therefore proposed an experiment to measure seed retention in a controlled environment across a large range of hydraulic conditions and vegetation architecture types. We used three types of artificial plants with contrasting morphotypes, and real seeds of Rumex crispus. Vegetation metrics were calculated on the basis of 3D plant models. We also tested the additivity of seed retention as a function of the length of vegetated area crossed by the seeds. We developed a semi-empirical formula for predicting seed retention. The main results of the experiment show that (i) the seed retention rate reacts differently to changes in density according to species, (ii) vegetation cover at the free water surface, potentially in contact with seeds, is a generic predictor of floating seed retention whatever the nature of the vegetated cover, and (iii) 95% of seed retention was reached for a large range of surface vegetation ratios and length of vegetation cover. The proposed formula could be used by stakeholders (farmers and ecologists) to estimate the amount of vegetation needed in a channel to limit or enhance seed dispersal.