Real life is often too complex to be understood by means of simple measurem
ents or simulations by models. The purpose of indicators is to simplify the
system so as to make this reality accessible to the users, in the form of
a diagnostic or decision aid tool. These indicators represent a compromise
between the scientific knowledge of the moment, the need to be concise, sim
plicity of use, and the availability of data. A procedure for developing su
ch indicators in 7 stages is proposed. These are always defined according t
o an objective. They are either measured, estimated, or calculated by aggre
gation of data. Their values are assessed relative to a reference. We propo
sed to evaluate their relevance by a "probability test." The validation of
indicators, i.e., the fact that they fulfil the objective for which they we
re intended, is done through a "usefulness test" by surveys of end-users. T
heir scientific value stems from the rigour of the procedure and the degree
of consensus which is established as to their method of elaboration. Even
when user-friendly models will be available, indicators will remain a privi
leged tool to understand complex systems, for example, to estimate the impa
ct of agricultural systems on the environment.