Reducing NOx emissions is one of the important indicators in the design of aeroengine combustion systems, and NOx estimation is one of the most important aspects in the design of emission reduction algorithms. Existing aero-engine NOx empirical models have limitations in their scope of application and not precise enough. In order to build a model that can estimate NOx emission accurately for aircraft engine, the chemical reactor network (CRN) model in Chemkin was constructed to characterize the combustion process of combustion chamber. The real data of NOx emissions were obtained based on the detailed chemical reaction mechanism. And the existing empirical model was checked. Pearson correlation matrix and control variable method were used to screen out characteristic parameters. The BP neural network prediction model of NOx was fitted and compared with applicable empirical models and actual conditions. The results show that under different thrust states, the maximum relative error between the prediction value of the constructed model and the true value is about 8.1%., The average relative error is 2.7% in the transition state, and both the results are smaller than those estimated by the empirical model. Therefore, compared with the traditional empirical model, the BP neural network prediction model constructed can predict engine NOx emissions more accurately.