Recent research has considered the phonological specificity of children’s word representations, but few studies have examined the flexibility of those representations. Tolerating acoustic–phonetic deviations has been viewed as a negative in terms of discriminating minimally different word forms, but may be a positive in an increasingly multicultural society where children encounter speakers with variable accents. To explore children’s on-line processing of accented speech, preschoolers heard atypically pronounced words (e. g. ‘fesh’, from fish) and selected pictures from a four-item display as eye movements were tracked. Children recognized similarity between typical and accented variants, selecting the fish overwhelmingly when hearing ‘fesh’ (Experiment 1), even when a novel-picture alternative was present (Experiment 2). However, eye movements indicated slowed on-line recognition of accented relative to typical variants. Novel-picture selections increased with feature distance from familiar forms, but were similarly sensitive to vowel, onset, and coda changes (Experiment 3). Implications for child accent processing and mutual exclusivity are discussed.