The goal of this research was to disentangle effects of phonotactic probability, the likelihood of occurrence of a sound sequence, and neighbourhood density, the number of phonologically similar words, in lexical acquisition. Two-word learning experiments were conducted with 4-year-old children. Experiment 1 manipulated phonotactic probability while holding neighbourhood density and referent characteristics constant. Experiment 2 manipulated neighbourhood density while holding phonotactic probability and referent characteristics constant. Learning was tested at two time points (immediate vs. retention) in both a naming and referent identification task, although only data from the referent identification task were analysed due to poor performance in the naming task. Results showed that children were more accurate in learning rare sound sequences than common sound sequences and this was consistent across time points. In contrast, the effect of neighbourhood density varied by time. Children were more accurate in learning sparse sound sequences than dense sound sequences at the immediate test point but accuracy for dense sound sequences significantly improved by the retention test without further training. It was hypothesised that phonotactic probability and neighbourhood density influenced different cognitive processes that underlie lexical acquisition. (Contains 2 tables and 2 figures.)