The paper presents a novel approach to predict the response of earthquake-excited
structures. The earthquake excitation is expanded in terms of series of deterministic
functions. The coefficients of the series are represented as a point in N-dimensional
space. Each available accelerogram at a certain site is then represented as a point in
the above space, modeling the available fragmentary historical data. The minimum
volume ellipsoid, containing all points, is constructed. The ellipsoidal models of
uncertainty, pertinent to earthquake excitation, are developed. The maximum response
of a structure, subjected to the earthquake excitation, within ellipsoidal modeling of
the latter, is determined. This procedure of determining least favorable response was
termed in the literature (Elishakoff, 1991) as an antioptimization. It appears that
under inherent uncertainty of earthquake excitation, antioptimization analysis is a
viable alternative to stochastic approach.