Three experiments investigated the relationship between memory for input and inductive learning of morphological rules relating to functional categories in a semiartificial form of Italian. A verbatim memory task was used as both the vehicle for presenting sentences and as a continuous measure of memory performance. Experiments 2 and 3 introduced increasingly explicit manipulations of attention to form compared to Experiment 1. In all experiments there were strong relationships between individual differences in memory for input as measured early in the experiment and eventual learning outcomes, and in Experiments 2 and 3 learning form-form (but not form-function) rules was related to vocabulary learning efficiency (taken as a measure of phonological long-term memory ability). These relationships along with the lack of an effect of feedback in Experiment 3 suggest that subjects tended to adopt a data-driven, as opposed to conceptually driven, mode of learning. However, the fact that the introduction of highlighting and vocabulary pretraining in Experiment 2 had a large impact on learning without improving early memory is taken to suggest that knowledge of distributional rules does not simply emerge out of memory encodings of the relevant forms but depends upon the appropriate allocation of attention over relationships between input elements at the time of encoding.