Abstract: To use data gathered during cigarette production, complex relationships between the influencing factors and the quality indexes of dried cut strips were studied through Bayesian network model analysis. The results showed that:1) The established network models could truly reflect the relationships and degrees of various factors on the quality indexes during cylinder drying process, and the results were consistent with the actual production outcome. 2) Under the controlled cylinder drying procedure, the most influential factors on the moisture content, the tobacco temperature and the moisture content in output tobacco from cooling procedure were the cylinder wall temperature in section Ⅱ, the opening of automatic valve in section Ⅱ and the cylinder wall temperature in section Ⅱ with the weight of 24.13%, 26.85% and 25.42% respectively. 3) The prediction accuracies of the established network models for the moisture content and the temperature of output tobacco from drying procedure were 83.14% and 82.67% respectively, both within the technical standard range. 4) During self-learning, compared with model M (using 806 batches of data), model N (using 1 261 batches of data) promoted the prediction accuracy by more than 1 percentage point for the moisture content in output tobacco from drying procedure and cooling procedure, and had a better self-learning ability.