We present a method for calculating the spectral albedo of snow which can be used at any wavelength in the solar spectrum and which accounts for diffusely or directly incident radiation at any zenith angle. For deep snow, the model contains only one adjustable parameter, an effective grain size, which is close to observed grain sizes. A second parameter, the liquid-equivalent depth, is required only for relatively thin snow. In order for the model to make realistic predictions, it must account for the extreme anisotropy of scattering by snow particles. This is done by using the “delta-Eddington” approximation for multiple scattering, together with Mie theory for single scattering. The spectral albedo from 0.3 to 5 μm wavelength is examined as a function of the effective grain size, the solar zenith angle, the snowpack thickness, and the ratio of diffuse to direct solar incidence. The decrease in albedo due to snow aging can be mimicked by reasonable increases in grain size (50–100 μm for new snow, growing to 1 mm for melting old snow). The model agrees well with observations for wavelengths above 0.8 μm. In the visible and near-UV, on the other hand, the model may predict albedos up to 15% higher than those which are actually observed. Increased grain size alone cannot lower the model albedo sufficiently to match these observations. It is also argued that the two major effects which are neglected in the model, namely nonsphericity of snow grains and near-field scattering, cannot be responsible for the discrepancy. Insufficient snow depth and error in measured absorption coefficient are also ruled out as the explanation. The remaining hypothesis is that visible snow albedo is reduced by trace amounts of absorptive impurities (Warren and Wiscombe, 1980, Part II).