A recursive Monte Carlo (RMC) method for estimating the importance function distribution in three-dimensional systems, intended for importance sampling applications, is developed. The method consists of dividing the system into relatively thin geometrical regions and solving the inhomogeneous forward transport equation for each of the regions. The RMC method is found to possess a number of unique features, including the ability to infer the importance function distributions pertaining to many different detectors from essentially a single Monte Carlo run. Various technical questions concerned with the practical application of the RMC method, including the questions of the accumulation of statistical and systematic errors and their dependence on the details of the system division and source batch size, are investigated. A promising algorithm for the application of the method is formulated. The practicality and efficiency of the RMC method is investigated for a number of monoenergetic problems.