A method was developed for early detection of dwarf off-types from micropropagated bananas (Musa sp., AAA Group, Cavendish subgroup). Selection of dwarfs was from glasshouse-grown plants, and although differences between dwarf and normal plants could be detected as early as 3 weeks from deflasking, discrimination was most effective at week 7, when the normal plants had reached a height of 18-20 cm. In order to develop selection criteria, known dwarf off-types and normal plants were micropropagated and established in a glasshouse. Measurements included plant height, petiole - - length, lamina length and width, and distance between leaves. Petiole length, lamina length, and the petiole to lamina length ratio provided the most promising selection criteria, with the dwarfs having significantly (P<0.01) smaller petioles and leaves than the normal plants. Selection was most effective when the plants were growing vigorously and uniformly. When growth became limiting, selection was more difficult. This was particularly apparent in plants that required repotting into larger containers and in micropropagated bananas grown in nurseries under suboptimal conditions. Plants were grown in the field and observed at bunch emergence, to verify trueness to type and to eliminate the possibility that off-types may have arisen in the normal and dwarf populations. A survey of these selection criteria with other Cavendish cultivars of various statures was also completed and the results suggest that dwarf and extra-dwarf off-types could be readily separated from the taller Cavendish cultivars Williams, New Guinea Cavendish, and Grande Naine.