Errors in understanding diagrams come from inappropriate ways of interpreting the diagrams, ways that have a basis in the visual structure of the diagram. In the case of information systems diagrams, these misconceptions can be diagnosed through a Bayesian causal network, in which latent misconceptions are inferred from a simple test on paths in diagrams. The misconceptions are related to surface errors in a merely probabilistic manner. Nonetheless, a model derived from one diagram can be used to make accurate predictions about errors on isomorphic diagrams. The technique can be used to assess misconceptions, including biases and bugs, and may be applied to many different problem domains. There are also pragmatic implications to this work: the domain of this application, that of a local area network, permeates information systems, and the diagnosis and correction of misconceptions will be helpful for those involved in information systems education, design, and trouble-shooting.