The physiological basis for differences in yields from well-watered and draughted plants of four contrasting clones of tea was studied in terms of light interception, dry matter production and partitioning at a high altitude site in Southern Tanzania where there are marked seasonal variations in rainfall and temperature. The plant dry weights, including roots, were measured eight months after field planting and subsequently at intervals of three to four months, corresponding to the different seasons, during the following two years. Fully irrigated plants of one clone (S15/10) were also harvested after four years in the field. Clones differed in the rates of canopy spread and hence in their capacity to intercept solar radiation. The ‘radiation use efficiency’ (the net total dry matter production per unit of intercepted short-wave radiation) was similar for the four well-watered clones and ranged from 0.40 to 0.66 g MJ−1, which corresponds closely to values reported for other woody tropical plants. A 16-week drought treatment imposed two years after planting reduced the mean light interception of the four clones by about 25% and the mean radiation use efficiency by 78% to 0.09 g MJ−1. Clone S15/10, a cultivar from Kenya which produces large yields, partitioned a greater proportion of dry matter to leaves and harvested shoots than the other clones, and correspondingly less to large structural roots. This resulted in a maximum harvest index of 24%, substantially greater than other values reported in the literature. There were seasonal differences in partitioning, with more dry matter being diverted to roots and less to shoots during the cool season. Although the drought treatments had no significant effect on root growth, the amount of dry matter partitioned to leaves, stems and harvested shoots declined by 80–95%. The roots of all four clones extended in depth at similar rates (about 2 mm d−1), those of Clone S15/10 reaching 2.8m after four years. The results are discussed in terms of appropriate field cultural practices and possible selection criteria for high yielding clones.