Résumé : In flowering plants, pollen dispersal is often the major contributing component to gene flow, hence a key parameter in conservation genetics and population biology. A cost-effective method to assess pollen dispersal consists of monitoring the dispersal of fluorescent dyes used as pollen analogues. However, few comparisons between dye dispersal and realized pollen dispersal have been performed to validate the method. We investigated pollen dispersal in two small populations of the insect-pollinated herb Primula elatior from urban forest fragments using direct (paternity analyses based on microsatellite DNA markers) and indirect (fluorescent dyes) methods. We compared these methods using two approaches, testing for the difference between the distance distributions of observed dispersal events and estimating parameters of a dispersal model, and related these results to dye dispersal patterns in three large populations. Dye and realized (based on paternity inference) pollen dispersal showed exponential decay distributions, with 74.2-94.8% of the depositions occurring at < 50 m and a few longer distance dispersal events (up to 151 m). No significant difference in curve shape was found between dye and realized pollen dispersal distributions. The best-fitting parameters characterizing the dye dispersal model were consistent with those obtained for realized pollen dispersal. Hence, the fluorescent dye method may be considered as reliable to infer realized pollen dispersal for forest herbs such as P. elatior. However, our simulations reveal that large sample sizes are needed to detect moderate differences between dye and realized pollen dispersal patterns because the estimation of dispersal parameters suffers low precision.