Study area The study was conducted in Bandarban district ofCHT, located in the southeastern part of Bangladesh. The topography ofthe region is mountainous. Two-thirds ofthe area is characterized by steep slopes; the remaining area comprises of undulating topography. Steep slopes combined with heavy seasonal rainfall pose serious limits to arable agriculture. A soil and land use survey conducted in 1966 found that 73% of the land of CHT is suitable only for forest, 15% for horticulture, and only 3% for intensive terraced agriculture (Brammer, 1986; Forestal, 1966). The agricultural land in the study area is broadly divided into two categories: sloping lands and valley lands. Valley lands, classified as Class-I land, are suitable for all forms of cultivation. Gentle sloping lands, classified as Class-II land, are suitable for terrace cultivation. Steep sloping lands, Class-III land, are considered to be only suitable for non-arable land use, like forestry and horticulture (Roy, 1995). Valley lands cover only 6% of the region. The soils on the hill slopes are dominantly brown loamy sands and silty clays; they are well-drained and strongly acidic (Gafur, 2001). While the soils on the river terraces are loams, soils in the valley floors are silty clay in texture. They are poorly drained in nature and subjected to seasonal flooding (ADB, 2001; Khan & Khisa, 2000). The climate is sub-tropical, with a mean monthly maximum temperature ranging from 25 to 34 v C; the annual average rainfall ranges between 2032 and 3910 mm, 80% of which occurs during the period May–September (ADB, 2001; Gafur, 2001). There are 11 hill tribes (Mongoloids) in Bandarban district, comprising more than half of the total population. The other people are Bengalese, who had migrated into the area from the neighboring plain region. Marma and Mro are the main tribal communities. The Marma normally live near streams and rivers; they are relatively better-off in their socio-economic condition than the Mro. The Mro usually live in higher elevations; they are the most disadvantaged of the tribal groups in the area. The Bawm, another tribal community, live on the hill slopes. These three groups account for about 80% of the total tribal population of the district. Agriculture is the main source of livelihood of these people, who mainly practice shifting cultivation. The crops traditionally used to grow under shifting cultivation are upland rice, maize, sweet potato, yam, vegetables, and spices such as chilly, turmeric, and ginger. The main cash crop is cotton, but the area is declining gradually. Non-farm income opportunities are very limited and in some areas non-existent. Data collection Primary and secondary information were collected from two representative subdistricts of Bandarban district, namely Bandarban Sadar and Alikadam. A total of 36 villages, 13 from Alikadam and 23 from Bandarban Sadar, were selected for the survey in such a way that it captured both ethnic diversity and locational variation to represent the characteristics of the entire CHT. The selected villages varied in terms of land use, accessibility, access to support services and facilities, and ethnic composition. Information on the establishment of the villages, distance to local markets and sub-district headquarters, time required to reach the road, existence of schools, educational facilities, the presence or absence of NGOs or missions and soil types, major land uses, and cropping patterns, were collected at the village level through semi-structured interviews with village headmen, discussion with villagers, observations and key-informants interviews. Biophysical information, including the altitudes of the villages, was obtained from topographical maps. Information about the areas under agriculture, crops cultivated, livestock, income from off-farm activities, and the educational attainment of the villagers was collected through a household survey. One-third of the households from each village were selected randomly for the household survey. Lists of households were obtained from the village headmen who update the lists every year for land-tax collection. The questionnaire was pre-tested and revised. Information regarding their service delivery systems and area coverage was also obtained from government agencies, NGOs and missionaries working in the area. Data analysis Twelve variables, namely proportion of area under shifting cultivation, horticulture, wetland cultivation, annual cash crops, total number of trees per household, number of fruit trees, number of timber species and number of cattle, pig, goat, poultry, and proportion of produce used for household consumption, were selected to characterize, classify and compare the existing land-use systems in the study area. ANOVA was applied to determine the significance of the selected variables. Table 2 shows the variables used in hierarchical cluster analysis and their statistics. The average linkages method of the cluster analysis with Pearson correlation was employed to generate clusters. In order to perform cluster analysis, the original variables were standardized as standard deviation of1. Collected information was analyzed using SPSS version 11. Qualitative data were transformed into quantitative through indexing for their use for multivariate analysis.