After viewing 27 upright photographs of faces, 94 subjects took a forced-choice recognition memory test in which the pairs were shown either upright (N = 54) or inverted (N = 40), then completed Marks' (1973) Vividness of Visual Imagery Questionnaire (VVIQ). Although groups of 12 good and 12 poor visualizers representing the lower and upper 30 per cent of VVIQ scores were less accurate, slower to respond and less confident for inverted than upright faces, VVIQ status did not interact with the effect of inversion. However, good visualizers were more confident than poor visualizers, and VVIQ scores themselves were lower (indicating more vivid reports) in the upright than in the inverted condition. It was also found that VVIQ scores were lower for more than for less confident subjects, but only for those who were less accurate. These results contradict the hypothesis that the VVIQ reflects holistic processing, but support the hypothesis that it is contaminated by an instrument factor. It is suggested that studies with the VVIQ should be designed to avoid cueing effects of the criterion task, and that the VVIQ should be accompanied by a test of general processing capacity to identify subjects whose responses might be contaminated by confidence.