Previous evidence suggests that the structure of similarity neighbourhoods in the developing mental lexicon may differ from that of the fully developed lexicon. The similarity relationships used to organize words into neighbourhoods was investigated in 20 pre-school children (age 3;7 to 5;11) using a two alternative forced-choice classification task. Children classified the similarity of test words relative to a standard word to determine neighbourhood membership. The similarity relationship between the test and standard words varied orthogonally in terms of type of similarity and position of overlap. Standard words were drawn from neighbourhoods differing in density. Results showed that dense neighbourhoods were organized by phoneme similarity in the onset+nucleus or rhyme positions of overlap. In contrast, sparse neighbourhoods appeared to be organized by phoneme similarity in the onset+nucleus, but manner similarity in the rhyme. These results are integrated with previous findings from infants and adults to propose a developmental course of change in the mental lexicon.