Fallen branches are a substantial component of coarse woody debris and a key ecological resource. The depletion of stocks of coarse woody debris since European settlement has contributed to the degradation of Australian grassy box woodlands, including the loss of biodiversity. Restoration options for remnant woodlands include the augmentation of coarse woody debris stocks. However, the extensive modification of grassy box woodlands has left few reference sites for establishing benchmarks to guide such restoration. In this paper we demonstrate a method for predicting fallen branch debris loads in the absence of reference sites, using data from a yellow box–red gum woodland. Our methodology is in two stages: first, the total volume of branch debris under individual trees was modelled; and second, these models were applied to groups of trees to predict stand-level loads of fallen branch debris. Although the models were developed for yellow box–red gum woodlands, the methodology would be applicable to other communities that lack reference sites. Predicted benchmark loads of fallen branch debris for yellow box–red gum woodland were between 7.0 m3 ha−1 and 11.9 m3 ha−1. Large senescing trees contributed the bulk of fallen branch debris. Model predictions indicated a 100-cm diameter at breast height (dbh) tree was 10 times more likely to produce debris than a 50-cm dbh tree, and if debris was present a 100-cm dbh tree produced approximately 10 times the volume of branch debris produced by a 50-cm dbh tree. These results highlight the importance of large senescing trees for the production of fallen branch debris and support the keystone role of large trees within remnant woodlands, and the need to conserve these structures. Our results also support the active management of regrowth woodland stands to facilitate the progression of individual trees to maturity and senescence. In particular, thinning of regrowth stands may promote the growth of retained trees, ensuring they contribute to fallen branch debris stocks with a minimum time lag.