OBJECTIVES: Diagnosing Appendicitis is a difficult challenge for emergency departments. Appendicitis has been the subject of numerous investigations, but a diagnostic tool of quality and applicability in the clinical routine has not yet been identified. We constructed a clinical predictive model for Appendicitis using a Bayesian network, using a decision analysis approach.  METHODS: We designed a methodology for the construction of decision models to support the process of clinical diagnosis, which was applied specifically to the diagnosis of Appendicitis. The methodology starts with a multiple correspondence analysis (MCA) for the selected variables with the highest power of discrimination, based on which an initial Bayesian network is proposed. This network was refined with the support of a group of medical experts in the diagnosis, to define the final structure of the Bayesian network. Finally, the model is validated through cross-validation method. For validation we used Sanabria, Bermudez, Dominguez and Serna’s (2007) database, which consists of 349 patients with suspected Appendicitis.  RESULTS: The implementation of the methodology produced results in a Bayesian network that included eleven variables associated to signs, symptoms and laboratory results. The results of the proposed model, with respect to medical assessment without specialized tools for diagnosis, showed significant improvements in accuracy, from 81% to 94%; in sensitivity, from 85% to 98%; and in specificity, from 76% to 90%.  CONCLUSIONS: The Bayesian network constructed based on the MCA improved the accuracy of Appendicitis diagnosis in comparison to other models available in the literature, among which are the Alvarado score and the Fenÿo score. Additionally, the constructed Bayesian network has characteristics that facilitate its applicability in the clinical routine.