The Risø Mesoscale PUFF model (RIMPUFF) is a Lagrangian atmospheric dispersion model that uses consecutive Gaussian puffs to simulate accidental release which is widely used in nuclear emergencies. Based on the real-time puff size and the distance between the puff and the monitoring station, RIMPUFF can calculate the dose results quickly by using the dose calculation interpolation table within its dose calculation module but the results are biased. To improve the accuracy of dose results under the requirement of rapid dose calculation, in this paper, a new interpolation table for dose calculation is proposed utilizing a 3D integral dose calculation method. And the proposed interpolation table was evaluated against the Belgian field experiment. In order to assess the quality of the optimization, the dose results of the new interpolation were compared with that of RIMPUFF, the 3D convolution method and observations from the field experiment. The comparisons indicate that the dose results of the new dose interpolation table are closer to the observed values and have better statistical metrics (0.39 for FB, 0.56 for NMSE, 0.72 for FAC2) than that of RIMPUFF (0.53 for FB, 0.76 for NMSE, 0.65 for FAC2).