Standard logical systems enjoy the properties of reflexivity, idempotence,
monotony. The latter property has been dropped in artificial intelligence s
ystems devoted to model drawing tentative conclusions. Up to now, all attem
pts to characterizing these so-called non-monotone systems are unsatisfacto
ry because these attempts either take absence of monotony as their major ch
aracteristics or give a collection of properties that also tolerate monoton
e systems. In contrast, we introduce in the context of non-constructive log
ics a characterization that indeed furnishes necessary and sufficient condi
tions for a system to be non-monotone. (C) 2000 Elsevier Science B.V. All r
ights reserved.