For literate individuals, does the spoken production of language involve access to orthographic codes? Previous research has rendered mixed results, with a few positive findings contrasting with a range of null findings. In the current experiments, we chose spoken Mandarin as the target language in order to better dissociate sound from spelling. Mandarin speakers named coloured line drawings of common objects with adjective-noun phrases (e.g., /lan2/ /hua1ping2/, blue vase). Adjectives and nouns were semantically and phonologically unrelated on all trials, but on critical trials they shared an orthographic radical. In two experiments, this resulted in a significant facilitatory effect on naming latencies. We interpret these results as strong evidence for the claim that retrieval of phonological codes in spoken production involves the co-activation of orthographic representations.