Two cloze-tests were constructed from written essays encoded by 80 first-year university students of middle-class and working-class origin. In a second experimental situation, 46 tertiary subjects were asked to ‘fill in’ the missing cloze deletions of these written passages. Within the terms of the Bernstein elaborated-restricted code framework it was posited that, since working class language is thought to be characterized by greater lexical and structural predictability, these passages would facilitate the decoding task. The analysis was based firstly on a ‘verbatim’ cloze completion criterion and secondly utilized an information theory approach. Results on the first criterion indicated significant social class differences (higher predictability of working-class messages on lexical and total cloze deletions); whereas those on the second criterion were nonsignificant. Possible implications of the study for teaching were explored.