Abstract There exists a large body of previous work using reduced two-dimensional models of the scrape-off layer (SOL), which model fluctuations in the drift-plane but approximate parallel transport with effective loss terms. Full size three-dimensional simulations of SOL turbulence in experimental geometries are now possible, but are far more computationally expensive than 2D models. We therefore use a flux-tube geometry model of the SOL to compare the results of 2D simulations to 3D simulations with a similar setup, looking for systematic differences. Our setup mimics the outer SOL of a double-null plasma, with no neutrals and no shear, and is therefore likely more favourable to agreement. Overall, we do find good agreement in the basic radial profiles, probability distribution functions, and power spectra of fluctuations. However, the average temperature is over-predicted in 2D relative to 3D, and we explain the difference in terms of the effect of geometrical simplifications of devices at low power. Varying geometric parameters, we find that supersonic flow in the divertor leg, which occurs because our simulations do not include neutrals and so represent low-recycling conditions, means that the divertor leg length only has a weak effect on the output. Finally, we examine the effect of altering the magnitude of source and sink terms in 2D, concluding that they cannot easily be used to recreate both the density and temperature profiles observed in 3D simultaneously.