Sampling data from a single cassava plant were used for testing various sampling methods. It was found that the degree of leaf damage was poorly correlated with the number of cassava green mites (CGM) present on a leaf. Sampling of specific mite stages should take into consideration the effect of leaf damage on the age distribution of the mites. Simple random sampling resulted in a high between-leaf variation caused by a vertical gradient in spider-mite abundance. The between-leaf variation could be reduced by stratified sampling. The optimum number of strata was three. Subsampling of leaflets combined with stratification provided the most precise estimates for a given sampling effort. Various regression methods for estimating the area of a cassava leaf from some simple measurements were compared. Combining the length of the longest leaflet with the number of leaflets gave the best prediction, explaining 95.8% of the total variation in observed leaf areas.