The processing of the uncertainty is one of the key problems in design expert systems for rockmechanics problems. Based on analysis of uncertainty, an uncertainty reasoning method processing ran-domness, fuzziness and incompleteness was proposed. Uncertainty propagation algorithms of single-pathand multi-path inference were discussed. The certainty values of knowledes and weights of various fac-tors were learned by neural networks. It offers automation of the kilowledge acquisition task associatedwith certainty values and weights, often the most difficult parts of knowledge extraction. This approachhas been applied to the expert systems developed by authors recentiy.The inlerence results show thatthis approach is scientific and practical.