Abstract: Nuclear fission data are important infrastructure data in nuclear applications and nuclear engineering. It is still challenging to obtain accurate and complete energy-dependent fission yields in experiments and theories. Bayesian Neural Network (BNN) is idea to treat inverse regression problems and can provide quantified uncertainties. We apply BNN to infer fission yields based on learning of existing fission yields. In particular, BNN is very useful for evaluations of fission yields when incomplete experimental data are available. We demonstrated that the BNN evaluations are quite satisfactory on mass distributions and energy dependencies of fission yields. This indicates that BNN is very promising in nuclear data community.