Sri Lanka is the world leader in made tea production and the small holding sector dominates national production by accounting for 60% of the island's tea production. However, given the high cost of production, there is a belief that it is very difficult to increase profitability without increasing costly inputs such as labour. With this background, in this study technical efficiency of the tea small holdings sector in the Mid Country Wet Zone of Sri Lanka was estimated in order to identify the potential to increase production without incurring any additional costs for inputs. The sources of inefficiency and the robustness of measured technical efficiency in various functional specifications was also investigated. The primary data collected during the period September - January 2001 relevant to sixty small holder tea producers in the Mid-country Wet Zone was used for the study. Maximum likelihood estimates of the stochastic frontier model were estimated for green leaf yield as a function of land extent, family labour, hired labour, fertilizer, chemicals, and dolomite, using Cobb-Douglas and translog models. The determinants of technical efficiency such as age of farmer, experience, education, occupation type of crop (VP/Seedling) and type of clone were investigated, following the Battese and Coelli (1995) specification. According to the Cobb-Douglas specification, extent of land, family labour, hired labour, fertilizer and dolomite showed significant effects on yield. The coefficients for land, family labour, hired labour and fertilizer had positive values of 1.11, 0.027, 0.067 0.029 and 0.004 respectively. The mean technical efficiency of the tea small holdings sector in the Mid Country Wet Zone was found to be 64.60 per cent. The result for the inefficiency model indicates that age of farmer, education, occupation, type of crop (VP or seedling) and type of clone have significant effects on efficiency. The coefficients for age, education, occupation and type of crop showed negative values. However contrary to expectations, type of clone and experience showed positive values. The estimation with the translog model yielded different technical efficiencies, which indicates the fact that technical efficiency estimations are highly sensitive to the functional form specified.