Aimed at the problems that the data are insufficient in manufacture, the risk factors are comparatively complex and the dynamic effect is significant, an improved model combined hierarchical holographic model (HHM) with Bayesian network (BN) is proposed. The risk identification framework of engine fuel accessories manufacturing is constructed by HHM. On the basis of this, the BN model is built, and its parameter learning method of maximum likelihood is improved. Finally, the effectiveness is verified by the example and Netica software. The results show that the total risk value of the engine fuel accessories manufacturing is 25%, and the key risk factors are dynamic balance and 302 hole lapping.