Abstract. Underground pipelines pose numerous challenges to 3D visualization. Pipes and cables are conceptually simple and narrow objects with clearly defined shapes, spanned over large geographical areas and made of multiple segments. Pipes are usually maintained as linear objects in the databases. However, the visualization of lines in 3D is difficult to perceive as such lines lack the volumetric appearance, which introduces depth perception and allows understanding the disposition and relationships between the objects on the screen. Therefore the lines should be replaced by volumetric shapes, such as parametric shapes (cylinders) or triangular meshes. The reconstruction of the 3D shape of the pipes has to be done on the fly and therefore it is important to select a 3D representation which will not degrade the performance. If a reconstruction method provides a good performance, the visualization of pipes and cables is guaranteed to provide a smooth experience to the final user, enabling richer scenes but also establishing the visualization requirements in terms of hardware and software to display underground utilities. This paper presents our investigations on a strategy for creating a 3D pipes for 3D visualisation. It is assumed that the pipelines are stored in a database and portions of them are retrieved for 3D reconstruction and 3D visualization. Generally, the reconstruction of underground utilities can be performed in different ways and should lead to realistic appearance, produce visual continuity between segments, include objects depicting specific connections and even consider buffer volumes displaying the uncertainty and the security distance between objects. The creation of such visually pleasing reconstructions may require very detailed shapes, which will increase the complexity of the scene and degrade the performance. This research has identified four criteria to measure the complexity of the scene and conclude on a 3D reconstruction strategy: number of scene graph nodes, number of triangles and vertices on the screen, needed transformations and appearance options. On the basis of these criteria a testing framework is developed. Ten different strategies for 3D reconstruction are defined and tested for X3D, X3DOM and WebGL. The paper analyses the results of the tests and concludes on the best strategy.