This research investigated whether automatic semantic priming is modulated by individual differences in lexical proficiency. A sample of 89 skilled readers, assessed on reading comprehension, vocabulary and spelling ability, were tested in a semantic categorisation task that required classification of words as animals or non-animals. Target words were preceded by brief (50 ms) masked semantic primes that were either congruent or incongruent with the category of the target. Congruent primes were also selected to be either high (e.g., hawk EAGLE, pistol RIFLE) or low (e.g., mole EAGLE, boots RIFLE) in feature overlap with the target. ‘Overall proficiency’, indexed by high performance on both a ‘semantic composite’ measure of reading comprehension and vocabulary and a ‘spelling composite’, predicted stronger congruence priming from both high and low feature overlap primes for animal exemplars, but only predicted priming from low overlap primes for non-exemplars. Classification of high frequency non-exemplars was also significantly modulated by an independent ‘spelling-meaning’ factor, indexed by differences between the semantic and spelling composites, which appeared to tap sensitivity to semantic relative to orthographic feature overlap between the prime and target. These findings show that higher lexical proficiency predicts stronger automatic semantic priming and suggest that individual differences in lexical quality modulate the division of labor between orthographic and semantic processing in early lexical retrieval.