A model of the child's learning of the past tense forms of English verbs is discussed. This connectionist model takes as input a present-tense verb and provides as output a past tense form. A new simulation is applied to 13 problems raised by critics of the model, presented as fundamental flaws in the conceptualizations underlying connectionism. The new simulation uses a new input representation based on feature/slot units and two views on input words, a new architecture using identity mappings and hidden units, a new learning algorithm, and an input corpus that includes all five cells of the English verb paradigm. Together the changes led to a vast improvement in the model's performance. All problems but those dealing with the issue of direct access within connectionist nets were addressed successfully. It is concluded that the connectionist models are extremely useful ways of characterizing the learning of inflectional systems, and that the critiques erred by confusing conceptualizations of the model with implementations. (MSE)