In this paper, an approach has been proposed in order to reconstruct the flat buildings using the LiDAR data and the digital aerial images because on one hand, these types of buildings constitute the main structure of IRAN mega cities. On the other hand, separating the roof planes and reconstructing them is a challenging matter due to the fact that the normal vectors of the building roof planes are completely the same and there is not any intersection among them. In this regard, firstly, 16 potentially primary features were produced and the optimum features were extracted using the genetic algorithm and the KNN algorithm to detect the buildings. Subsequently, an approach was presented to eliminate the misclassified regions and to improve the detection results. In the designed reconstruction approach, each building parcel was considered separately in order to reduce data redundancy and increase the result accuracy. After selecting the considered parcel, an initial class for building roof planes was achieved by employing the surface slope differential. Actually, a threshold was specified and the regions with the slope differential value less than it were removed. Therefore, an initial class of primary planes were recognized and labeled by connected component algorithm. The roof planes were improved and detected completely by textural and altitudinal analyzing. The acceptable range was determined by computing the median minus the variance of the elevation in the plane to the median plus the variance. The median was selected as a criterion, because it is not sensitive to the noise of data and it causes to choose a reliable value. To identify the adjacent planes, the recognized plane was scanned row by row and column by column. In each row/column, the pixels with values more than zeros were extracted and analyzed. If there exists a variation, the pixels numbers were extracted and considered as the adjacent. Afterwards, the boundary nodes of each plane were extracted using the chain code algorithm. The optimum nodes should be selected as boundary and the planes should be placed beside each other without any intersection and gap. Hence, by investigating the distance and angle, very close nodes were removed and replaced by their mean. Also, the nodes that cause creating the intersection or gap were recognized and rectified in order to eliminate this error. Then by searching around of the extracted nodes, the corresponding terrain node for each boundary node was obtained. The equation of each plane was computed by the coordinates of the inner nodes in that plane. Finally, the equation of planes, extracted boundaries, and nodes of the floor were utilized for reconstructing the final model. The proposed approach was implemented on some building blocks with different structures and the accuracy of the reconstructed plane was evaluated in both altimetric and planimetric criteria. The evaluated results were shown 84.56% accuracy on average for planimetric reconstruction, 0.212 meter root mean squared error for planimetric corner coordinate, and 0.145 meter root mean squared error for altimetric reconstruction. These results clarify the good performance of the proposed approach for reconstructing buildings with flat roofs.