The height estimation of every tree included in NFI sample plots is based on height-diameter ratio measurements of small amount of subsample trees. Height is measured for systematically selected every seventh tree on the sample plot (Kuliešis et al. 2003a, 2009b). On an average, height and diameter of 1–7 trees of prevailing and 1–3 trees of admixture tree species are measured per one plot. Heights of the remaining trees are estimated using generalized height-diameter model (Kuliešis 1993) based on measured diameters of all trees as well as height-diameter ratio of subsample trees. The goal of this study is to estimate the accuracy of tree height prediction using generalized height-diameter model based on measurement of height-diameter relation of small amount of subsample trees. This study makes use of the data of height-diameter ratio measurements on 1998–2010 NFI plots. This data derives from 47.5 thousand subsample trees of prevailing eight tree species as follows: pine, spruce, birch, aspen, black alder, grey alder, oak, and ash, and 24.6 thousand subsample trees of the same tree species as admixture in stands. Deviations of modeled heights from measured ones and their significance were estimated in respect of tree species, tree position – prevailing or admixture, mean diameter, and mean height. Spruce and oak stands have the most, and black alder and aspen stands have the least tree height variation per group with the range of mean diameter ± 2 cm and the range of mean height ± 1 m. Accuracy of mean tree height estimation per plot shifts from 3–5% in stands with mean diameter equal to 40 cm and more up to 5–15% in stands with mean diameter equal to 4–20 cm. Mean height deviations of predicted values from those obtained by measuring all tree populations on stand is negligible for all species, -0.6 cm per prevailing tree, -2.5 cm per admixture tree and -1.3 cm per any tree. Differences between predicted and measured height both of prevailing and admixture trees of all species are statistically insignificant (probability level 0.95). Tree species were divided into two groups with different character of height deviations for trees with diameter lower or larger than mean diameter of stand. Systematical deviations up to 9–20 (33) cm were estimated for predicted heights of trees with diameter up to and over the mean diameter in pine, oak, birch and black alder stands, while in spruce, aspen, grey alder and ash stands such deviations had random character not exceeding 6–9 (24) cm. While all the system in general (sampling design and model) used for tree height estimation during NFI can be given a positive evaluation, some height estimation improvements for prevailing pine and spruce trees less than 7 m high and admixture birch trees less than 5 m high should be done by modifying the generalized height-diameter model applied.

**Key words:** generalized height-diameter model, subsample trees, predicted heights, accuracy