This paper presents a system for autonomous information gathering in an inf
ormation rich domain under time and monetary resource restrictions. The sys
tem gathers information using an explicit representation of the user's deci
sion model and a database of information sources. Information gathering is
performed by repeatedly selecting the query with the highest marginal value
. This value is determined by the value of the information with respect to
the decision being made, the responsiveness of the information source, and
a given resource cost function. Finally, we compare the value-driven approa
ch to several base-line techniques and show that the overhead of the meta-l
evel control is made up for by the increased decision quality.