Methods of generating turbulent in ow boundary conditions for LES and DNS are surveyed and ranked according to their effectiveness. Methods included in the survey are random number pseudo turbulence, rescaling of isotropic turbulence data in space and/or time, Taylor's hypothesis applied to a frozen three-dimensional turbulent velocity field, and the direct transfer of turbulent velocity data from an auxiliary simulation. The simpler approaches are shown to be acceptable for flows with strong inviscid instabilities such as distorted isotropic turbulence or turbulent wakes. These same approaches are found to fail in wall-bounded flows where weak viscous instabilities prevail. Wall-bounded flows appear to require the more sophisticated methods, which are largely based on the direct transfer of accurately-simulated turbulent data. Effcient means of generating such data are also surveyed. Very simple time-dependent, simulation-based inflow generation procedures such as a parallel-flow boundary layer are shown to yield a quantum leap in accuracy over the simpler methods, including those that apply Taylor's hypothesis to a accurately-simulated frozen velocity field. While the parallel-flow boundary layer method is adequate in many cases, greater accuracy can be achieved if necessary by accounting for spatial growth effects via Spalart's original method, or by more recent variants. These rather sophisticated methods are still cost effective and can also account for the effects of pressure gradients.