An improved algebraic iterative algorithm was proposed to solve the problem that reconstruction of gas diffusion can barely be completed by the conventional algebraic iterative algorithm. Under the condition of fan beam geometry and double projections, the prior matrix was introduced in the iteration steps and the results processed by median filter every few iterations. Taking Gaussian plume model as test object, the effect of iterations, relaxation factor and different filter using strategies on the reconstruction results were analyzed. The experimental results shows that the mean square error and peak error reach the basic convergence and the peaks can be controlled in a certain range when the number of iterations is 500 and the relaxation factor is 2. Compared with the conventional algebraic iterative algorithm, the improved algorithm is better on the reconstitution indicators of mean square error and peak error. Furthermore, the peak position can be located more accurately and tail peak can be reconstructed successfully through the improved algorithm and the improved algorithm is robust to noise.