A probe recall short-term retention task was used to test the applicability of the "phonological recoding" (Conrad, 1972) and "flexible decoding" (Smith, 1972) models to processing tactics used by readers of Chinese. Subjects were 45 adult speakers of Cantonese. Stimuli were lists of Chinese characters which varied in terms of phonological and visual similarity. The dependent variable was the proportion of correctly recalled target characters from "recent" positions. Significant primary memory interference effects were observed for phonologically similar lists but not for visually similar lists. Results were interpreted with reference to the two models of skilled reading. (Author)