The two main models of children's acquisition of inflectional morphology--the Dual-Mechanism approach and the usage-based (schema-based) approach--have both been applied mainly to languages with fairly simple morphological systems. Here we report two studies of 2-3-year-old Polish children's ability to generalise across case-inflectional endings on nouns. In the first study, we found that the morphological form in which children first encounter a noun in Polish has a strong effect on their ability to produce other forms of that same noun. In the second study, we found that this effect is different depending on the target form to which children are switching. Similarity between inflectional endings played a crucial role in facilitating the task, whereas the simple frequency of either source or target forms was not a decisive factor in either study. These findings undermine Dual-Mechanism models that posit all-or-none acquisition of abstract morphological rules, and they also present serious challenges for usage-based models, in which frequency typically plays a key role. (Contains 5 figures, 5 tables and 8 footnotes.)