A common and important practice in reservoir characterization is to classify a reservoir into certain fundamental elements or facies, because knowing the facies generally improves the petrophysical correlations that are used in engineering calculations. Facies classification is usually done by visual inspection of core or patterns in wireline log responses. Either way, the classification must be extended to other observations, usually to previously unclassified wireline logs, that are not part of the training set. This paper presents a novel method for making this extension based on the generation of probability logs constructed from Bayes theorem. A probability log is the probability of the existence of a particular facies at a specific position in a well. It is constructed from combinations of wireline logs that have been previously classified into facies using an elementary form of Bayes theorem. The process allows a priori identification of which logs are the most useful in facies prediction. Compared to neural networks, the Bayesian procedure is very simple and uses much less effort. In fact, we use a spreadsheet to generate probability logs. The same procedure can be used to identify net pay from a log determined porosity cutoff. This paper gives the details behind the Bayesian procedure. We generate probability logs for two types of reservoirs, an offshore sandstone and a West Texas carbonate, and compare the accuracy of the predictions to those obtained from neural networks. The accuracy (percentage of classifications predicted correctly) is nearly identical to that obtained from neural networks, but effort expended in training is far less.