Basic density (BD), fibre length (FL), fibre width (FW), and fibre wall thickness (FWT) were investigated in 46 Norway spruce (Picea abies (L.) Karst.) trees from five different stands in eastern Norway. From each tree, wood samples were collected in different radial and longitudinal positions. Random coefficient mixed models were used to investigate variation within as well as among trees, both within and among stands. The R-2 with random effects included, describing the best possible (individual) fit of the observed data to the models, were 0.90 for BD, 0.99 for FL, 0.88 for FW, and 0.91 for FWT. With only fixed effects, the best model explained 56% of the total variation for BD, 94.5% for FL, 61% for FW, and 63% for FWT. A common model for all trees, without tree and site information, predicted FL well but BD, FW, and FWT poorly. Adding site index, breast height diameter, and tree height to the models reduced the residual variance considerably for FW, FWT, and particularly BD, whereas only a minor improvement was gained for FL. The latter type of models might be easier to use for industrial purposes. Although information about ring width gave further improvements, ring width measurements are time consuming and difficult to perform in the forest and in industrial environments.
