This paper uses a stochastic frontier and inefficiency model to test the efficiency of grape production in the Western Cape. The data covers two panels of wine grape farms (34 in Robertson and 36 in Worcester) for 2003 and 2004 and 37 table grape farms in De Doorns for 2004 only. Tests show that Cobb Douglas stochastic production frontiers, with variables to explain the inefficiencies are an appropriate representation of the five individual samples. The stochastic frontier results indicate that output can be explained by land, labour and machinery and that efficiency cab be affected by labour quality, age and education of the farmer, location, the percentage of non-bearing vines and expenditures on electricity for irrigation. These data is sufficiently good to produce reasonable results without pooling, but most applied economists would consider the possibility of improving the estimates by pooling the samples. However, pooling tests show that in this situation with small samples, when pooling is permissible it may not be helpful. More effort on determining the true distributions is needed to improve the way such samples are handled and Bayesian methods may be helpful in this respect.