Forests are complex ecological systems, characterised by multiple-scale structural and dynamical patterns which are not inferable from a system description that spans only a narrow window of resolution; this makes their investigation a difficult task using standard field sampling protocols. We segment a QuickBird image covering a beech forest in an initial stage of old-growthness – showing, accordingly, a good degree of structural complexity – into three segmentation levels. We apply field-based diversity indices of tree size, spacing, species assemblage to quantify structural heterogeneity amongst forest regions delineated by segmentation. The aim of the study is to evaluate, on a statistical basis, the relationships between spectrally delineated image segments and observed spatial heterogeneity in forest structure, including gaps in the outer canopy. Results show that: some 45% of the segments generated at the coarser segmentation scale (level 1) are surrounded by structurally different neighbours; level 2 segments distinguish spatial heterogeneity in forest structure in about 63% of level 1 segments; level 3 image segments detect better canopy gaps, rather than differences in the spatial pattern of the investigated structural indices. Results support also the idea of a mixture of macro and micro structural heterogeneity within the beech forest: large size populations of trees homogeneous for the examined structural indices at the coarser segmentation level, when analysed at a finer scale, are internally heterogeneous; and vice versa. Findings from this study demonstrate that multiresolution segmentation is able to delineate scale-dependent patterns of forest structural heterogeneity, even in an initial stage of old-growth structural differentiation. This tool has therefore a potential to improve the sampling design of field surveys aimed at characterizing forest structural complexity across multiple spatio-temporal scales.