Benson & Perrett's (1991 b) computer-based caricature procedure was used to alter the positions of anatomical landmarks in photographs of emotional facial expressions with respect to their locations in a reference norm face (e.g. a neutral expression). Exaggerating the differences between an expression and its norm produces caricatured images, whereas reducing the differences produces 'anti-caricatures'. Experiment 1 showed that caricatured (+50% different from neutral) expressions were recognized significantly faster than the veridical (0%, undistorted) expressions. This held for all six basic emotions from the Ekman & Friesen (1976) series, and the effect generalized across different posers. For experiment 2, caricatured (+50%) and anti-caricatured (-50%) images were prepared using two types of reference norm; a neutral-expression norm, which would be optimal if facial expression recognition involves monitoring changes in the positioning of underlying facial muscles, and a perceptually-based norm involving an average of the expressions of six basic emotions (excluding neutral) in the Ekman & Friesen (1976) series. The results showed that the caricatured images were identified significantly faster, and the anti-caricatured images significantly slower, than the veridical expressions. Furthermore, the neutral-expression and average-expression norm caricatures produced the same pattern of results.