The increased availability of 3D urban data reflects a growing interest in 3D spatial analysis. As 3D spatial analysis often uses complex 3D data, studies of the potential gains of using more detailed 3D urban databases for specific uses is an important issue. First, more complex data implies an increase in time and memory usage for the analysis (and calls for more research on the efficiency of the algorithms used). Second, detailed 3D urban data are complex to produce, expensive and it is important to be well informed in order to decide whether of not to invest in such data. Currently, many studies have been led about the fitness for use of 2D data but they are very scarce concerning 3D data. This article presents a method to determine the influence of 3D modeling on the results of 3D analysis by isolating the potential sources of errors (such as roof modeling and geometric accuracy). This method is applied on two 3D datasets (LOD1 and LOD2) and a 3D indicator (the sky view factor or SVF). The results show that the significant influence of roof modeling is globally compensated by the difference in geometric modeling but that important local variations are noticed. Nevertheless, for 75% of the SVF processed the difference between the results using these two databases is lower than 2%.