Karst aquifers are complex and linked to numerous management issues, particularly concerning the the risk of flooding or the availability of water resources. In view of the associated stakes, it is necessary to have the best possible knowledge of the hydrodynamic behavior of these aquifers. In this context, a semi-automated unsupervised clustering of the hydrodynamic behavior at different piezometers has been developed. It was applied to the Cadarache site of the the French Alternative Energies & Atomic Energy Commission (CEA) in South-East of France; site which is subject to rapid groundwater floods. At the methodological level, this paper proposes a grid for the interpretation of sorted water levels in relation to the hydrogeological properties of the formations. The results of this study, achieved on 75 boreholes, allows identifying several clusters of boreholes. Three clusters appeared to be representative of specific areas. The method only uses sorted graphs of water levels. A relevant relation between each cluster and its geological structure is highlighted. This study revealed the role of preferential flow paths of NW-SE and NE-SW lineaments previously identified by regional tectonic or structural studies and with same directions than fractures observed at local scale (borehole core).