When developing real-world applications of Bayesian networks one of the largest obstacles is the highly time consuming process of gathering probabilistic information. This paper presents an efficient technique applied for gathering probabilistic information for the large SACSO system for printing-system diagnosis. The technique allows the domain experts to provide their knowledge in an intuitive and efficient manner. The knowledge is formulated in terms of likelihoods, calling for methods to transform it into conditional probabilities suitable for the Bayesian network. The paper outlines a general transformation method based on symbolic propagation in a junction tree.