This paper introduces an approach to detect and 3D reconstructs buildings using aerial imagery and LiDAR data. This research consisted of three phases, building detection, 2D building outline reconstruction and 3D building reconstruction. In phase building detection, firstly off-terrain objects including trees and buildings are extracted from LiDAR data. Secondly Support Vector Machines (SVMs) Algorithm is employed to differentiate trees and buildings. Training data which are used in SVMs are choose in semi-automatic procedure. After eliminating trees, K-means clustering algorithm is used to separate buildings which are not in same elevation. Results are showing our building detection method was successful in detection of small and large buildings. Completeness, Correctness and Quality for building detection results respectively are 86.60%, 99.10% and 85.92%. In phase 2D building outline reconstruction, firstly, the building boundaries have been vectorized. Then produced boundaries are generalized and unnecessary line segments are removed. After generalization, a new approach is used to build orthogonal buildings. In this research, 3D building reconstruction is done in LOD2. For detection of roof structure of buildings, the parameters of plane that have been fitted to LiDAR points within each kernel is obtained. With considering these parameters as features of each building, ISO-Data clustering has been done. The results of this clustering represent the planar surfaces of each building. So, a plane is fitted to each class (planar surface) with least squares method. Then, within the boundaries of each building, roof patches which have similar plane parameters and are close together are merged. Plane parameters of integrated roof patches again are determined. Finally, 3D models of buildings have been reconstructed with intersection of planar surfaces and obtaining of vertex points of each building. Elevation and total RMS values of specified planes for roof structure of buildings respectively are 0.4 m and 0.9 m.