A study of 13 sample plots (0.01 ha) in 6 forests showed that the average number of topsoil samples required per plot to secure 95 per cent confidence limits for a range about the mean of + l o per cent was 6 for total nitrogen, 9 for total phosphorus and 29 for 0.5 M acetic acid extractable nutrients (calcium, magnesium, phosphorus and potassium). In some plots 95 per cent confidence intervals, based on 5 composite samples. for extractable nutrients were as large as the range of mean values for different plots within a forest. High variability within plots causes large variation in correlation coefficients between tree growth and soil properties so that the intensity of sampling should be of the order indicated above to identify factors affecting growth. For predicting timber yield the accuracy of regression equations containing predictors of high within-plot Variability is too low to be of practical value. The sampling effort required to achieve a given level of precision does not increase greatly when plot size is increased from 0.01 to 0.1 ha. Since Yield Class, the most useful measure of growth, is not designed for areas less than 0.5 ha, the use of the larger plot size is recommended.