The estimation of seismic loading requires both the featuring of the seismotectonic environment as well as a prognosis of site-specific strong ground-motion parameters. Due to the long recurrence times of larger events, there is a lack of high quality strong motion recordings at many sites. The use of seismograms obtained from different seismotectonic regions, additionally recorded under different subsoil conditions, poses several questions as to the significance of the assessment of seismic loading. A way to overcome these problems is the generation of synthetic seismograms, e.g. by the use of Boore's stochastic procedure. Subsoil conditions may be explicitly included in the simulation as proposed by Kunze et al. In this context, the seismotectonic environment forms the deterministic frame of any stochastic strong ground-motion simulation. In terms of Boore's approach, seismic moment, global stress drop, high frequency cut off, and focal depth act as controlling parameters. In order to achieve a more realistic assessment of possible seismic loading, a classification of seismoactive zones should account for parameter vectors as described above rather than for single parameters like magnitudes or intensities. A tool to attack this task may be found in ‘cluster analysis’, which is a type of pattern recognition based on statistical considerations. Crucial to this technique, however, is the use of properly estimated parameters, which still cause discussion in the seismological community. Problems and chances of the approach are discussed with examples from the Swabian Jura and the Rhine graben.