Four experiments with faces support the original interpretation of categorical perception (CP) as only present for familiar categories. Unlike in the results of Levin and Beale (2000), no evidence is found for face identity CP with unfamiliar faces. Novel face identities were shown to be capable of encoding for immediate sorting purposes but the representations utilized do not have the format of perceptual categories. One possibility explored was that a choice of a distinctive face as an end-point in a morphed continuum can spuriously produce effects that resemble CP. Such morphed continua provided unequal psychological responses to equal physical steps though much more so in a better likeness paradigm than for forced-choice recognition. Thus, researchers doing almost the same experiments may produce very different results and come to radically different conclusions.

Photographs of morphed faces were shown to close friends of portrayed individuals. Three tasks were used: localisation of a morphed target on the continuum between the two original faces, simultaneous same - different discrimination of face pairs separated by a 20% morphing step (AB task), and sequential classification of the same pairs (ABX task). Localisation data were plotted against morph coefficients. Evidence of categorical processing was provided by steeper functions for upright vs upside-down faces. In the AB task, intermediate faces were discriminated better than faces separated by the same morphing step but closer to one original. This was confirmed in a control experiment where the participants were unfamiliar with portrayed individuals and were unlikely to process our stimuli categorically. The superiority of intermediate faces in the AB task was attributed to a nonlinearity of continua generated by the morphing procedure, and used as a baseline to evaluate ABX classification data. Also in the ABX task, intermediate faces, those straddling the categorical boundary, were classified more accurately than faces located on the same side of the boundary. However, the superiority in classification accuracy was larger than the superiority in discrimination accuracy operationalised by the AB task, as predicted by the categorical perception hypothesis.