The problem of site dependent estimation of the seismic strong motion for Sofia region is discussed. Deterministic modelling for micro-zonation of Sofia region is implemented for receiving an estimation of local site conditions as result of displacements provoked by seismic events in Sofia region with stochastic and neural modelling. An approach for seismic waves classification on the base of principal axis transformation, long-range dependence time series analysis and neural modelling is suggested. With the help of scene oriented model are determined the boundaries of destructive phase of strong motion acceleration. For selected diapason of destructive phase is implemented two-dimensional vector quantization for transformed accelerograms with maximal and middle values. With self-organizing map are determined weight centres of selected classes of destructive phase. Learned and trained self-organizing map optimize selected target classes and determine probability density function. Suggested model helps to provide side dependent estimation of the seismic strong motion for Sofia region.