Twenty white British and 20 black African students were subjects. Ten white European male faces and 10 black African male faces were randomly selected from a larger set of color full-face portraits taken with a Polaroid camera. Each subject was required to encode, or construct from memory, one white and one black face using the Caucasian and Afro-Asian versions of the Photofit Kit respectively. To provide accuracy indexes, 2 different groups of 20 white decoders matched the resulting Photofits to photographs of the original 20 faces. Photofits made of white faces by white subjects were more accurately decoded than were those of white faces made by black subjects. No difference was found between the black and white encoders in the accuracy with which they constructed Photofits of black faces. Both white and black encoders made significantly better constructions of white faces compared with black faces. The lack of racial difference in encoding black faces underlined the paucity of alternative facial features contained in the Afro-Asian Photofit Kit compared with those offered in the Caucasian kit. On average, there are about two Caucasian features for one Afro-Asian feature. Another explanation for the results is that the Photofit was originally developed for the reconstruction of white faces and therefore biased the manner in which the faces were segmented. The inescapable conclusion of the study is that Photofits made of black faces are likely to be inferior to those made of white faces regardless of the color of the witness. Acknowledgements, tabular data, and nine references are given.