Estimates of forest and woodland structure prior to major periods of modification (e.g. prior to European settlement) are routinely used to inform decisions relating to biodiversity conservation, silviculture and carbon sequestration potential in natural forests and woodlands. The techniques used to derive these estimates often demand that data be collected from specific geographic locations (e.g. locations for which historic survey records exist, where pollen accumulates, or where there is little modification by humans since European settlement) and therefore are often inherently biased. In this study we predicted numbers of trees by diameter class for several widespread forest and woodland types in south-eastern Australia under conditions of relatively little modification by humans since European settlement. To do this we fitted Generalised Additive Models (GAMs) separately to counts of stems in eight diameter classes from 495 plots using explanatory variables representing human modification, environmental variation and natural disturbances. We predicted stem densities under conditions of relatively little modification by humans since European settlement from these models by holding the significant explanatory variables representing modification by humans at minimum observed values.We compared these predictions with published estimates of pre-European stem densities and estimates that we derived for stands at theoretical equilibrium using the quotient of diminution (q). Our mean predictions were broadly comparable with estimates derived from both of these sources; however, we appeared to over-estimate numbers of stems in the smaller diameter classes for some vegetation alliances. A key outcome of this research-and rarely reflected in other techniques used to predict pre-European forest structure â¿¿ was the amount of variation in stem numbers even within a single diameter class and vegetation alliance. For example, in the white box vegetation alliance, flat parts of the landscape supported 5 times the number of large trees (>80cm DBH) found on upper slopes under conditions of relatively little modification by humans since European settlement. Our results therefore suggest that these forests and woodlands are more structurally heterogeneous than typically reflected in pre-European estimates and vegetation alliance should not be the unit for managing these stands. The methodology we present is applicable in many forests where the objective is to predict forest structure under conditions of relatively little modification by humans.