It is well known that in the general context the similarity relation i
s very fuzzy and hard to define. Unfortunately, the intuitive notion o
f similarity is not a transitive relation: knowing that A is similar t
o B and that B is similar to C does not necessarily imply similarity b
etween A and C. This is a main obstacle when trying to express formall
y what a coherent font design is. In this paper we suggest a method to
decompose complex letter forms into simpler elements and we suggest a
formal transitive definition of a similarity relation between these e
lements. In the context of digital typography, this definition enables
developing an algorithm to recover classes of similar elements within
different characters of a given font. This knowledge is further explo
ited to ensure coherent type processing. For example, a modification (
e.g. by a type designer) of a character element is propagated automati
cally to all the other characters that include a similar element. For
the moment, the discussion is limited to the class of stroke fonts. (C
) 1997 Elsevier Science Ltd. All rights reserved.