The unfalsified control concept provides a theoretical basis for directly v
alidating closed-loop controller performance from open-loop experimental da
ta, or even from closed-loop data acquired while another controller was in
the feedback loop. It allows this to be accomplished without making any pri
or assumptions about the plant or noise. This approach to experimental cont
roller validation avoids the conservatism inherent in indirect two-step val
idation approaches that require one to first identify a plant model and unc
ertainty-bounds from experimental data before analysing control system robu
stness. Controller validation is shown to be the feedback generalization of
open-loop model validation concepts that form the basis of various control
-oriented system identification theories. Copyright (C) 2001 John Wiley & S
ons, Ltd.