Kemler Nelson (1984) reported an experiment (Expt 4) intended to demonstrate that the processing of categories evolves with development from a holistic to an analytic mode. Five‐ and 10‐year‐old children were asked to learn two categories of schematic faces which were either defined by a single attribute value or structured by overall similarity. As the identification of the children's processing modes was ambiguous, we replicated this experiment, with additional criteria to differentiate between holistic and multi‐attribute processing. Results showed that both the 5‐ and 10‐year‐old children used an analytic processing mode. The main developmental difference concerned the number of attributes on which their categorization judgements were based. These results can be related to the format o