Factors affecting the accuracy with which adults could assess the age of unfamiliar male faces aged between 5 and 70 years were examined. In the first experiment twenty-five 'young' adult subjects, aged 16-25, and twenty-five 'old' adults, aged 51-60, were used. Each subject saw five versions of three different faces: these consisted of an original version of each face and four manipulated versions of it. The manipulations consisted of mirror reversal, pseudo-cardioidal strain, thresholding, and elimination of all but the internal features of the face. The second experiment was similar except that a between-subjects design was used: each subject saw three faces for each age category of target face, but was exposed to only a single type of manipulation (plus a set of 'original' faces which were identical for all groups, so that the comparability of the different groups in age estimation could be checked). Results from both experiments were similar. Age estimates for unmanipulated 'original' faces were highly accurate, although subjects were most accurate with target faces that were within their own age range. Results for the manipulated faces implied that the importance of cardioidal strain as a necessary and sufficient cue to age may have been overestimated in previous reports: subjects' age estimates were accurate when cardioidal strain was absent from the stimulus, and poor when cardioidal strain was the only cue available.