Although it is commonly assumed that biomass allometric models are site specific, evaluations of site-effects are rarely undertaken. In this paper we develop biomass-allometric models to determine site influences. This study is based on data from 240 Norway spruce trees (Picea abies (L.) Karst.), growing in 24 early-growth plantations. A multilevel modelling approach was adopted and intraclass correlation was used to evaluate site effects. Results indicated that biomass allometric models were highly specific to sites and that, depending on the biomass component and the type of predictor adopted, some 33% and 86% of overall model variance could be attributed to forest stand effects. The remaining variance was attributable within stand variability. Stem biomass was the most site-specific biomass component whereas branch biomass was the least influenced by site effects. Diameter at collar height (D) was less site-specific than height (H) in predicting biomass. Using D and H within the same model as distinct predictors, although improving the model fit, increased the model site-specificity. However, when D and H were combined in one predictor expression (i.e. (DH)-H-2), this reduced model site specificity, despite requiring fewer parameters than other models. This also compensated for undesirable collinearity effects amongst predictor variables. Furthermore, for the sampled diameter range, the site-specificity was mainly driven by biomass allocation pattern (to branches, needles and roots). The considerable between site variability of allometric relationships suggests that consideration of stand effects is essential for the robust prediction of biomass.