The formability of aluminium alloy sheet is affected by its microstructure, including the crystallographic texture and its impact on anisotropy. However, obtaining a description of sheet anisotropy, such as a yield surface, requires a prohibitively large number of mechanical tests. In principle, crystal plasticity simulations calibrated on a reduced number of tests, making use of experimental measurements of texture and grain shape, can be run in place of these mechanical tests to determine yield surfaces. Although several such workflows have been proposed, it is still not evident how successful this approach is for commercial aluminium alloys. In addition, it is not clear what kind of crystal plasticity model, (e.g. full-field, mean-field) is required, what the optimal method of sampling the experimentally measured texture is, or to what extent microstructural aspects like grain shape affect the quality of the predictions. In this work, we present a reproducible computational workflow which uses stress-strain data from mechanical tests and texture data from electron-back scatter diffraction (EBSD) measurements to calibrate material parameters, simulate deformation, and fit a yield surface. Model predictions of anisotropy were validated by comparison with experimental results. Sensitivity studies comparing full-field and mean-field crystal plasticity methods were carried out, in which the effects of model resolution and texture sampling method were studied. The aim was to determine the applicability of the crystal plasticity virtual testing method in the calculation of yield surface for aluminium alloy sheet, and to determine the most efficient method of carrying out this calculation with a view to using it in an industrial context. It was found that while the approach works, the distribution of second phase particles affected the mechanical anisotropy in a way unaccounted for by the crystal plasticity models. Additionally, more experimental comparisons are required to determine whether the full-field predictions outperform the less computational demanding mean-field predictions.