This paper explores the hypothesis that children pay more attention to phonological cues than semantic cues when acquiring grammatical patterns. In a series of artificial allomorphy learning experiments with adults and children we find support for this hypothesis but only for those learners who do not show clear signs of explicit learning. In particular, learners who cannot verbalize a correct rule after the experiment nevertheless perform above chance on phonological patterns, but not the semantic ones. On the other hand, learners, particularly adults, are more likely to (explicitly) discover and successfully verbalize a rule based on a salient feature of animacy compared to a phonological feature based on the number of syllables. We discuss implications of these results in the context of a distinction between explicit and implicit learning mechanisms and how this distinction relates to the study of phonological bias.