Forested landscapes are an important component of many large virtual environments in games and film. In order to reduce modelling time, procedural methods are often used. Unfortunately, procedural tree generation tends to be slow and resource-intensive for large forests. The main contribution of this paper is the development of an efficient procedural generation system for the creation of large forests. Our system uses L-systems, a grammar-based procedural technique, to generate each tree. We algorithmically modify L-system tree grammars to intelligently use an instance cache for tree branches. Our instancing approach not only makes efficient use of memory but also reduces the visual repetition artifacts which can arise due to the granularity of the instances. Instances can represent a range of structures, from a single branch to multiple branches or even an entire tree. Our system improves the speed and memory requirements for forest generation by 3–4 orders of magnitude over naïve methods: we generate over 1;000;000 trees in 4.5 seconds, while using only 350MB of memory.