Availability of operational regional hydrodynamic models and near real time Harmful Algal Bloom (HAB) alerts from monitoring stations and remote sensing products have allowed the proliferation of short term advective HAB forecasts. However, their predictive ability in simulating HAB transport needs to be continuously evaluated in events of different HAB species to assess their applicability to different domains and the impacts of the choices made in model setup. Here we review the performance of three different modelling systems which were part of the PRIMROSE project against historical bloom events in different regions in the European Atlantic Area. The objectives are to understand their predictive ability and to demonstrate some aspects of Lagrangian model setup that are relevant to HAB early warning systems; in particular the use of advection-diffusion only models (without a biological component) and the effects of model configuration, especially model resolution. Hindcast and forecast simulations have been run in examples of high biomass blooms detected in satellite imagery; in the western English Channel, several events of potentially toxic species like Karenia mikimotoi and Prorocentrum cordatum (minimum) were simulated and in Western France a bloom of Mesodinium rubrum, prey of the toxic Dinophysis spp. Additionally, some simulations for studying the evolution of low biomass Dinophysis spp. blooms in Galicia-North Portugal were undertaken with models of different setup. Several metrics have been used to quantify the model performance and to compare the results of the different model configurations, showing that differences in hydrodynamical model configuration (initiation, resolution, forcing, and simulation domain) result in differences in the predicted transport of HABs. We find that advection only is a reasonable approximation but that it may do worse in an early (onset) phase than later on, and we find transport is generally increases with increasing resolution. Our results confirm that Lagrangian particle tracking tools can be integrated operationally in HAB early warning systems providing useful information on potential HAB evolution to users.