To estimate oil yield in research trials, fresh fruit bunch (FFB) yield is multiplied by the oil/bunch determined on a sample of bunches, but there is no standard method for sampling bunches. Data from palms in which all bunches were analysed for seven years allowed actual mean oil/bunch (O/B) and oil yields to be calculated, and compared with estimates from samples. Samples of five bunches per palm gave poor precision, with 95% confidence limits of about ± 4% O/B. Increasing the number of bunches to 10 per palm gave confidence limits of about ± 2.5% O/B. Alternatives include analysing more bunches only if the coefficient of variation (CV) for five bunches is large, or measuring fruit/bunch and oil / mesocarp on 10 bunches, and mesocarp / fruit, which is less variable, on only two or three bunches. The mean O/B of five bunches may give acceptable estimates of oil yield per palm. For progeny or treatment mean oil yields, the mean of 50 bunches per progeny gives a CV of about 10%. Precision is increased if a bunch is sampled from every palm, rather than at random from the progeny. When considering mechanisation of bunch analysis, a method in which the bunch components are determined on different bunches has no disadvantages.