Diagnosis is currently a very important problem in process automation. In recent years, artificial neural networks have been used in many diagnosis applications. In this paper, two artificial neural network models are proposed to solve the diagnosis problem. The first network involves the application of the back propagation learning algorithm. The second model involves the application of the category learning algorithm. As a case study, they are applied to diagnose liver diseases. Both proposed models have been extensively studied to reach the optimum values of network parameters. A wide range of learning examples has been chosen to cover all classes of input with different degrees of uncertainty. Both models have proved to be capable of learning. The backpropagation model has learned after 13000 iterations with a minimum of five processing elements in the hidden layer, whereas the category-learning model needed only 1000 iterations with a minimum of ten processing elements in the hidden layer. The tuning of parameters to give accurate representation of the input classes was more difficult in the category-learning model. Both proposed models have proved to give accurate diagnosis even if the networks are given an incomplete set of inputs or if the inputs are applied to the network with varying degrees of uncertainty. In spite of the difficulties in the learning phase of the category learning model, it gives more conclusive diagnosis results