In view of the problems that the look-up table interpolation method used in aircraft fuel measurement is low in efficiency, low in accuracy, and is not good at the fault tolerance of the neural network applied to the calculation of aircraft fuel quantity, the fuel quantity algorithm based on RBF neural network is studied. By enhancing the discrete distribution of the fuel tank volume characteristic database to optimize the training samples, refining the neural network training algorithm to improve the fault tolerance for the input data, and employing genetic algorithm to optimize design parameters of the neural network, generalization capability and training efficiency of the RBF neural network in the fuel quantity calculation are effectively improved. According to the calculation example of an aircraft fuel tank and corresponding ground tests, the data dispersed method of tank models in this paper can further accurately describe their volume characteristic, with a 34.8% reduction in RMSE of interpolation calculation compared to the equidistant cutting method. The developed RBF neural network is good at the calculation accuracy, improving efficiency compared with the interpolation calculation method being about 5 times; Compared with the OLS algorithm, the improved algorithm has a 61.5% reduction in the estimated RMSE of test samples when the input parameters have errors, and the fault tolerance is significantly improved; The proposed method has a certain of practical value in engineering.