Abstract: In this paper we consider the task of refining a low-resolution radiosity solution by using final gather. This final gather can be done by using Monte-Carlo methods. We show how to built an adequate probability density function pdf such that the involved variance is greatly reduced and thus we can reduce the number of samples to take, keeping the error. The proposed pdf uses information gathered during the computation of low-resolution radiosity. This data includes an approximation to the distribution of irradiance landing on a patch and coming from other patches. Once this distribution is (approximately) known, we can use it to built an importance-based pdf for final gather, such that just a few tens of samples can be taken to accurately compute irradiance.