Monte Carlo method is used to simulate the radiation transport problems which have characteristics of high temperature, high pressure, multi-media and large deformation of the grid. Due to the nonuniform particle distribution and great difference in the sizes of the grid, it makes the statistical errors of the grid fluxes fluctuate wildly. The error accumulation after many time steps may even influence the truth of the calculated results. So the global source biasing technique is developed. The importance function used here for source biasing sampling is based on the grid fluxes and errors of the previous time step. This is much faster than the traditional methods for solving the adjoint equation. The numerical tests showed that the neutron flux errors in the important grids are obviously reduced by using global source biasing which adaptively assigns the number of particles in each grid.