This talk will describe the SHDOM algorithm for computing radiative transfer in three-dimensional cloudy atmospheres. SHDOM is the most widely used deterministic (i.e. non-Monte Carlo) model used for 3D transport in atmospheric physics. It uses a source function iteration approach in which the source function is represented by a spherical harmonics series at each grid point, while integration of the source function to compute intensity is carried out along discrete ordinates. The choices made in the algorithm design to maximize computational speed and minimize memory use will be described. The advantages and disadvantages of SHDOM compared to Monte Carlo methods will also be discussed.