Thee owpasttheRoyalAircraftEstablishment (RAE)2822airfoilisstudiedbythree-dimensionalcomputational e uid dynamics (CFD), which includestheviscous side walls of thewind-tunnel test, with an aspect ratio of3. Three turbulence models, two aspect ratios, and several grids are used, as well as two treatments of the e oor and ceiling (neither one actually representing slotted surfaces ) and small Mach number and incidence adjustments. The results deviate from two-dimensional results sufe ciently to revisit the quantitative conclusions and the ranking of turbulence models that were made from two-dimensional CFD in the 1980s and 1990s. However, contrary to our hopes, the three-dimensional effects fail to improve the pressure recovery after shock-induced separation in the more dife cult case 10, so that all of the turbulence models we tried still fail to match measurements by modern standards,evenwithMachnumberand angle-of-attackadjustments.Theunseparated case6producesvery similar trends. With present levels of computerpower, tests with fully documented three-dimensional solid-wall boundary conditions appear most desirable, but axisymmetric test cases can already be quite useful.