Much previous research on implicit learning has examined form-based sequential regularities over letters and syllables. Recently, however, researchers have begun to examine implicit learning of systems in which the regularities are described at the level of meaning. We review existing work in this area, primarily from vision research and natural language. These studies suggest that meaning-based generalisations can be learned without intent and without awareness of what those generalisations are. In the case of language we review work on learning semantic constraints on determiner usage, and the acquisition of semantic preferences of verbs. We discuss outstanding issues: whether noticing of meaning, as well as form, is necessary, whether the effects reflect learning of new form-meaning connections as opposed to tuning of existing ones, and whether some semantic distinctions are more available to the implicit learning process than others.