The growing amount of 3D point cloud data obtained by airborne laser scanning, terrestrial laser scanning and image matching, allows a variety of different calculations and data analysis. Applications range from monitoring tasks (e.g. structural health monitoring, landslides monitoring, etc.), archaeological evaluations, vegetation mapping to 3D city modeling. For processing huge data sets with millions of points highly efficient algorithms are required. Segmentation of point cloud data provides a grouping of points based on a similarity criterion. This group information enables efficient access to points with the same properties. Thus segmentation is one of the first steps within the processing chain of many applications. This thesis presents a concept for segmentation of large point clouds with Seeded Region Growing. Since the processing unit can not read huge dataset into main memory, the data must be divided into smaller parts. The point cloud is divided into rectangular non-overlapping parts (tiles). The tiles are then processed independently within the segmentation. This allows parallel computation by distributing tiles to multiple processing threads. Afterwards adjacent segments from different tiles are merged. As it is shown the results of the segmentation do not depend on the tile size, but are mainly influenced by the similarity criterion. The point cloud can thus be divided into arbitrary tiles to optimize for processing speed and memory footprint.