In-vitro preparations of the human lung combined with high-resolution tomography can be used to derive precise models of the human lung. To develop an abstract graph representation, specially adapted image processing algorithms were applied to segment and delineate the bronchi. The graph thus obtained contains topological information about spatial coordinates, connectivities, diameters and branching angles of 1453 bronchi up to the 17th Horsfield order. The graph was analyzed for statistical and fractal properties and was compared with current models. Results indicate a model that exhibits asymmetry and multifractal properties. This newly established reference model is an important step forward in geometrical accuracy of the bronchial tree representation that will improve both analysis of lung images in clinical imaging and the realism of functional simulations.