The seismic behavior of massive geotechnical systems exhibits complex response patterns and mechanisms under severe loading conditions. Some of these mechanisms are localized in space, but nevertheless impact significantly the entire system response and ultimately its stability. A thorough monitoring and identification of the whole response of a distributed soil system commonly constitutes a significant challenge, and would generally be prohibitively expensive. This study presents a point-wise identification technique of soil-systems using the acceleration records provided by local instrument arrays. The newly developed identification algorithm consists of: (1) evaluation of strain tensor time histories using the motions recorded by a cluster of instruments (arranged in an appropriate multi-dimensional configuration), (2) estimation of the stress tensors corresponding to the evaluated strains utilizing a pre-selected class of constitutive models of soil response, (3) computation of accelerations associated with estimated stresses using the equations of motion, and (4) calibration and evaluation of an optimal model of soil response by minimizing discrepancies between recorded and computed accelerations. Computer simulations and analysis of centrifuge tests of a soil -quay wall system showed that the proposed technique provides an effective tool to identify local soil characteristics and properties.