A method of assisting failure diagnosis in a system includes obtaining data including a probabilistic Bayesian Network describing a set of failures, a set of symptoms and probabilities of at least some of the symptoms being associated with at least some of the failures in a system. A cost value representing a cost associated with learning of a presence or absence of the symptom is obtained for at least some of the symptoms, as well as a plurality of information values, e.g. values representing measures of information gained by learning of the presence or absence of the symptom in relation to a respective plurality of the failures, associated with the symptom. The method then computes an information-for-cost value for the symptom based on the cost value and the plurality of information values.