This paper examines the problem of system identification from frequenc
y response data. Recent approaches to this problem, known collectively
as 'Estimation in H-infinity', involve deterministic descriptions of
noise corruptions to the data. In order to provide 'worst-case' conver
gence with respect to these deterministic noise descriptions, non-line
ar data algorithms are required. In contrast, this paper examines 'wor
st-case' estimation in H-infinity when the disturbances are subject to
mild stochastic assumptions and linearity in the data algorithms is e
mployed. Issues of convergence, error bounds, and model order selectio
n are considered. (C) 1998 Elsevier Science Ltd. All rights reserved.