There are three challenging issues in traditional seismic facies classification based on seismic attributes.Firstly,it is difficult to introduce priori-information into the processing of classification to enhance the result of seismic facies classification.Secondly,it is difficult to quantitatively evaluate reliability of the result for seismic facies classification.Thirdly,it is difficult to determine the weights of all parameters of Bayesian networks in classification.In order to solve the above-mentioned problems,this paper proposes a new approach of seismic facies classification based on Bayesian networks,which effectively combines the priori-information and probability distribution of the training samples to construct a reasonable classification model,and deduce the probability for each of seismic facies.According to the probability distribution of each seismic facies,we could estimate the reliability of the classification results in a quantitative manner.The principles and workflow are presented in detail for applying Bayesian networks to seismic facies classification.The numerical experiment proves that this method is correct and feasible.