From a data set encompassing the years 1990-2008 pairs of surface drifters with maximum initial separations of 5, 10 and 25 km have been identified. Model trajectories have been calculated using the same initial positions and times as the selected pairs of surface drifters. The model trajectories are based on the TRACMASS trajectory code and driven by the ocean general circulation model NEMO. The trajectories are calculated off-line, i.e. with the stored velocity fields from the circulation model. The sensitivity of the trajectory simulations to the frequency of the stored velocity fields was tested for periods of 3 and 6 h as well as 5 days. The relative dispersion of the surface-drifter and model trajectories has been compared, where the latter was found to be too low compared to the relative dispersion of the drifters. Two low-order trajectory sub-grid parameterisations were tested and successfully tuned so that the total amplitude of the relative dispersion of the model trajectories is similar to that associated with the drifter trajectories. These parameterisations are, however, too simple for a correct simulation of Lagrangian properties such as the correlation time scales and the variance of the eddy kinetic energy. The importance of model-grid resolution is quantified by comparing the relative dispersion from an eddy-permitting and a coarse-resolution model, respectively. The dispersion rate is halved with the coarse grid. The consequences of the two-dimensionality of the trajectories is evaluated by comparing the results obtained with the 2D and the Lagrangian 3D trajectories. This shows that the relative dispersion is 15% stronger when the trajectories are freely advected with the 3D velocity field.