Current overemphasis on nosological diagnosis has led to a neglect of the process of symptom recognition. There is evidence, however, that the perception of the symptom alone does not guarantee symptom ascertainment since a decision-making component is also involved. To achieve the latter, additional information must be provided by the contextual cues implicit in the ongoing diagnostic hypothesis. Current diagnostic systems, however, still assume a two-stage model according to which symptom and disease recognition are independent cognitive events. This paper suggests that this model is inadequate and that descriptive psychopathology is not transparent. It then describes a neural network simulation to make various aspects of the problem explicit. This takes into account the multidimensional and probabilistic aspects of symptom recognition and is, from this point of view, superior to traditional algorithmic models. It also has the capacity to represent the different cognitive styles involved in symptom recognition.