It was difficult to develop a normal parametric model to match the results well for the complex non-linear relationship among the inlet concentration,retention time and removal efficiency of hydrogen sulfide in the biotrickling filter.Aiming at this problem,back-propagation neural network(BPNN) was introduced into the research.A non-parametric model was established by the training of BPNN with the experimental data.And this model could simulate and predict the experimental results well.The correlation coefficient of simulation with all experimental data was 0.986 0.In nine different experimental conditions,the linear regression correlation coefficient of network predicted values and the corresponding experimental results was as high as 0.965 9.