This paper presents a hierarchical case representation using a context guid
ed retrieval method. The performance of this method is compared to that of
a simpler flat file representation using standard nearest neighbour retriev
al. The estimation of construction costs of light industrial buildings is u
sed as the test domain. Each case in the system has approximately 400 featu
res. These are structured into a hierarchical case representation that hold
s more general contextual features at its top and specific building element
s at its leaves. Problems are decomposed into sub-problems and solutions re
composed into a final solution. Comparative results show that the context g
uided retrieval method using the hierarchical case representation is signif
icantly more accurate than the simpler flat file representation using stand
ard nearest neighbour retrieval. (C) 1998 Elsevier Science B.V. All rights
reserved.