Preselection of rules is an algorithm which chooses one rule from linguisti
c description of some process. This rule is then used for inference instead
of the whole description. This article studies properties of preselection
and proposes use of measures of resemblance for choosing the rule which bes
t fits the observation.