A field experiment is reported concerning ten one-week sensitivity training groups. Groups were composed-according to their scores on Schutz's FIRO-B. It was predicted that the optimal conditions for change were provided by heterogeneous compositions. Three groups were composed to be heterogeneous for control behaviour but homogeneous for affection behaviour. Three further groups were homogeneous for affection but heterogeneous for control. The remaining four groups were heterogeneous for both need areas. Changes on Shostrom's Personal Orientation Inventory were found in all conditions. Control-heterogeneous groups increased on 'existentiality' and 'nature of man' subscales, while affection-heterogeneous groups increased on 'inner-directedness' and 'feeling reactivity'. Groups heterogeneous for both need areas showed change on all four of these POI scales. Most change thus resulted from the groups with the most heterogeneous compositions.