Three experiments investigating children's performance on tests of sound categorization are reported. In two experiments, the children had to decide which was the "odd one out" in a sequence of three, four, and five words. In each case the "oddity" judgement required selecting the word which differed in its final consonant, producing a different rime segment. The odd word differed either by one phonetic feature (voice or place) or by both voice and place, from the others in the sequence. Oddity judgements were easier when the odd word was phonetically dissimilar to the background items but, contrary to expectation, performance did not decline as list length increased. Experiment 3 extended these findings by showing that place-only changes were more difficult than voice-only changes. Sound categorization performance correlated with memory span for words, but memory span did not make an independent contribution to sound categorization once age and reading skills had been controlled. The experiments show that the sound categorization task taps individual differences in speech perception mechanisms but that memory load is less crucial to this task than has often been supposed.