This paper addresses the relationship between the frequency of morphological, lexical and syntactic forms in the input and children's language acquisition. The problem of obtaining reliable frequency measures under different sampling regimes is discussed. Since children are not simple associationist processors onto which the frequency of hearing or producing a string maps directly, a number of factors that interact with frequency are reviewed: form-function mappings, neighbourhood relations and multiple cues. These factors raise the problem of the level of granularity at which we test for a frequency-based explanation which can only be established through empirical research. Studies showing a relationship between the relative frequency of forms in the input and children's errors, including morphological errors, optional infinitive errors and accusative-for-nominative errors in English are discussed. The final section of the paper deals more briefly with some counter-arguments to the importance of frequency effects in the learning of grammar. © 2010 Elsevier B.V.