Comparisons of inflow conditions for large-eddy simulations of turbulent, wall-bounded flows are carried out. Consistent with previous investigations, it is found that the spectral content of the inflow velocity is important. Inflow conditions based on random-noise, or small-scale eddies only, dissipate quickly. Temporal and spatial filtering of a time series obtained from a separate calculation indicates that it is important to capture eddies of dimensions equal to or larger than the integral length scale of the flow. Three methods for generating inflow velocity fields are tested in a simulation of spatially developing turbulent channel flow. Synthetic turbulence generation methods that introduce realistic length scales are more suitable than uncorrelated random noise, but still require fairly long development lengths before realistic turbulence is established. A recycling method based on the use of turbulent data obtained from a separate calculation, in different flow conditions, was found to result in more rapid transition. A forcing method that includes a control loop also appears to be effective by generating turbulence with the correct Reynolds stresses and correlations within less than ten channel half heights.