The objectives of the study were to assess changes in fine root anisotropy and specific root lengths throughout the development of Eucalyptus grandis (W. Hill ex Maiden) plantations and to establish a predictive model of root length density (RLD) from root intercept counts on trench walls. Fine root densities (<1 mm in diameter) were studied in 6-, 12-, 22-, 28-, 54-, 68-and 72-month-old E. grandis plantations established on deep Ferralsols in southern Brazil. Fine root intercepts were counted on 3 faces of 90-198 soil cubes (1 dm 3 in volume) in each stand and fine root lengths (L) were measured inside 576 soil cubes, sampled between the depths of 10 cm and 290 cm. The number of fine root intercepts was counted on one vertical face perpendicular to the planting row (N t ), one vertical face parallel to the planting row (N l ) and one horizontal face (N h ), for each soil cube sampled. An overall isotropy of fine roots was shown by paired Student's t-tests between the numbers of fine roots intersecting each face of soil cubes at most stand ages and soil depths. Specific root lengths decreased with stand age in the upper soil layers and tended to increase Plant Soil (2010) 334:261-275 in deep soil layers at the end of the rotation. A linear regression established between N t and L for all the soil cubes sampled accounted for 36% of the variability of L. Such a regression computed for mean N t and L values at each sampling depth and stand age explained only 55% of the variability, as a result of large differences in the relationship between L and N t depending on stand productivity. The equation RLD= 1.89*LAI*N t , where LAI was the stand leaf area index (m 2 m −2 ) and N t was expressed as the number of root intercepts per cm 2 , made it possible to predict accurately (R 2 =0.84) and without bias the mean RLDs (cm cm −3 ) per depth in each stand, for the whole data set of 576 soil cubes sampled between 2 years of age and the end of the rotation.