Computation solutions for the flow around a cube, which were generated as part of the Computational Wind Engineering 2000 Conference Competition, are compared with full-scale measurements. The three solutions shown all use the RANS approach to predict mean flow fields. The major differences appear to be related to the use of the standard $k-{varepsilon}$, the MMK $k-{varepsilon}$ and the RNG $k-{varepsilon}$ turbulence models. The inlet conditions chosen by the three modellers illustrate one of the dilemmas faced in computational wind engineering. While all modeller matched the inlet velocity profile to the full-scale profile, only one of the modellers chose to match the full-scale turbulence data. This approach led to a boundary layer that was not in equilibrium. The approach taken by the other modeller was to specify lower inlet turbulent kinetic energy level, which are more consistent with the turbulence models chosen and lead to a homogeneous boundary layer. For the $0^{circ}$ case, wind normal to one face of the cube, it is shown that the RNG solution is closest to the full-scale data. This result appears to be associated with the RNG solution showing the correct flow separation and reattachment on the roof. The other solutions show either excessive separation (MMK) or no separation at all (K-E). For the $45^{circ}$ case the three solutions are fairly similar. None of them correctly predicting the high suctions along the windward edges of the roof. In general the velocity components are more accurately predicted than the pressures. However in all cases the turbulence levels are poorly matched, with all of the solutions failing to match the high turbulence levels measured around the edges of separated flows. Although all of the computational solutions have deficiencies, the variability of results is shown to be similar to that which has been obtained with a similar comparative wind tunnel study. This suggests that the computational solutions are only slightly less reliable than the wind tunnel.