Increased use of multi-phase, wood-based biocomposites may contribute to sustainable development. The porous microstructure offers unique possibilities for modification, but global properties are often predicted based on simplified unit cells and homogenization. For materials design, simulations based on complex 3D microstructures with statistical variability are alternatives to better understanding physical properties. Parametric models are developed in a distortion-map-based method to represent 3D wood microstructures. Basic structures of uniform tubular cells and other features are generated followed by distortion mapping. These maps are highly adaptable and can generate realistic features and variability. Fibers, vessels, and ray cells are realistically distributed. The models are realistic, versatile, and scalable, as well as can be used to simulate the mechanical, optical, and hydrodynamic properties of complex composites. The model is promising for generating large sets of data to train deep learning networks for multi-physics research.