In this paper we show that the classical distance between fuzzy relations i
s not adequate when it is used to test expert systems. We introduce a new o
ne that solves the drawbacks detected and that reduces the computational co
st. We also study the synthesis of diagnosis that is used by the distance i
ntroduced. We finish testing an expert system according to the method intro
duced. (C) 2000 Elsevier Science B.V. All rights reserved.