the variability of forest soil properties and the number of samples required to achieve desired levels of precision for estimation of property means have received little attention in the tropics. highly variable forest soil properties require more intensive sampling and often have less predictive value for site assessment purposes. the study also compared soils of natural forests and areas converted to teak (tectona grandis linn. f) plantations (21.3 ± 5.1 years) in the offinso and juaso forest districts in the ashanti region, ghana. sites selected for this study were in the moist semi-deciduous forest zone and had nearly identical physiographic characteristics. a simple random sampling procedure was used to obtain soil samples at each site. in each of three natural forest stands and three teak plantations, 16 soil pits were examined and soil samples from the 0-20 (major rooting depth) and 20-40 cm depth were analysed for selected chemical and physical properties. in the 0-20 cm depth, coefficients of variation varied from 8% (ph) to 72% (available p), and in the 20-40 cm depth from 16% (ph) to 116% (available p) under teak plantations. similarly, in the 0-20 cm depth coefficients of variation varied from 11% (ph) to 40% (exchangeable k) and in the 20-40 cm depth from 10% (bulk density) to 86% (available p) under natural forests. under both cover types, more samples were required to estimate means at ±10% allowable error with a confidence level of 95% for chemical properties than for physical properties.