The present paper concerns systems that memorize concepts. More specif
ically, we consider two models of the connectionist paradigm. The firs
t one is the classical ''schema-model''. The second one is a model tha
t has been proposed by the present author. The schema-model has some p
sychologically useful properties, but it has also some drawbacks. The
second model includes an additional layer and it uses a different type
of distributed representation: Due to these elements, it is possible
to overcome the drawbacks at issue and to obtain a model that describe
s more psychological facts.