Applications for places in UK Higher Education are usually made before the results of A-level examinations are known, so references from schools and colleges normally refer to expected (or predicted) grades. Inaccuracies in these predictions may be systematically related to key characteristics of the applicant and could lead to under-representation from various groups of students. This paper examines data on predicted A-level grades for 415 recently-enrolled university students. In contrast to the findings of previous studies however, we find that prediction bias is not particularly related to the gender, class or schooling of the student, but is closely linked to the predicted grades themselves--students predicted low grades performed above expectations, and vice-versa. The implications of this for current UK government initiatives intended to widen participation in Higher Education are considered briefly in the conclusion. (Contains 4 tables, 2 figures and 15 notes.)