A lexicon can be viewed as network of words connected by phonological similarity relations. In such a network model, words that sound similar are connected to be neighbours. The most often used criterion for defining neighbours is the one-phoneme difference rule (Luce & Pisoni, 1998), which states that any two words that only differ in one phoneme by addition, deletion or substitution are considered as phonological neighbours. Previous studies have shown that children’s phonological neighbourhoods are structurally different than those calculated from the adult lexicon (Charles-Luce & Luce, 1990; Coady & Aslin, 2003, among others). However, we also know that by the time children fully acquire the language, their phonological neighbourhoods should be similar to those in the adult lexicon. Thus the question becomes how and when does the structural shift take place during language development. As far as we know, this issue hasn’t really been addressed in the literature. The current study intends to fill in the gap by presenting a developmental analysis of the phonological neighbourhoods in two children’s lexicon. The two children, Joel and Ruth, both in their second and third year of life during the study, were selected because their data presented interesting contrast in a previous study on grammar development. Our current results show that the structural properties of these two children’s early phonological neighbourhoods are comparable to the findings in previous studies. More detailed network analyses suggest that they use different strategies for acquiring words in different stage. In the very early stage of acquisition, they tend to acquire words that are maximally different from each other but do not necessarily turn out to be structurally important later. Later, after their phonological neighbourhoods have reached a certain level of maturity, they tend to use a different strategy which prefers to acquire words that are structurally important first.