This paper describes the 3D formulation of MR-FDPF algorithm (multi resolution frequency domain parflow) for indoor propagation prediction for centimetric waves. Usual approaches developed for indoor propagation are based either on empirical or ray-tracing techniques. The former suffers a lake of accuracy while the later offers the possibility of a trade-off between accuracy and computational load. But in severe environments ray tracing like models have difficulties to predict correctly multiple diffraction and reflections that are dominant. The discrete method we propose here is similar to the famous TLM approach in the frequency domain. In the frequency domain, the problem, that is in fact a linear system resolution, can be solved within a multi-resolution framework. The originality of our work is the way that the propagation is computed over a multi-resolution binary tree architecture, referring encapsulated nodes on which the propagation is recursively computed. The full 3-D implementation of MR-FDPF is described here, and a few results are presented. Our algorithm is well adapted to complex environments because all reflections and diffraction are taken into account.