In this chapter we present importance-driven subdivision techniques for efficient computing of radiosity. The purpose of these techniques is to perform an intelligent selection of interesting directions from a patch, based on viewpoint complexity. The first technique uses the importance of a direction to approximately compute form factors by adaptive subdivision of a hemi-cube. The second technique uses a heuristic function, which estimates viewpoint complexity in a given direction from a patch and recursively subdivides a hemisphere. This technique permits to improve the classical Monte Carlo progressive refinement technique by choosing good directions to send rays. The third technique is an improved variant of the second one. The heuristic function used to estimate the visual complexity in now applied to a region delimited by a pyramid and the visibility criterion used by the function is more accurate and easier to compute. A fourth technique, based on more accurate estimation of viewpoint complexity, is also proposed. This technique seems interesting but it is not yet implemented.