The polyphasic approach to taxonomic problems has led to the creation
of complex datasets that lend themselves to numerical analysis. The nu
merical study, however, has to deal with mathematical problems linked
with the presence of mixed-type data originating from the investigatio
ns. Correspondence analysis (CA) is an ordination technique widely use
d in ecology and social sciences but only rarely applied to taxonomic
problems. In Cb corresponding variables and taxa ordination are obtain
ed simultaneously, thus allowing to explore the taxonomic interrelatio
nships between taxa and variables in a single analysis. CA can be used
on large and small datasets, and can be applied to mixed-type data af
ter appropriate coding. It is not sensitive to variation of class numb
er and size and is useful to screen large unstructured datasets, to su
ggest which variables should be retained to discriminate samples, to d
etect outliers or erroneous data and to perform identification of unkn
own samples. It also has the advantage of handling missing data partic
ularly well. On the other hand, CA is sensitive to outliers, which can
cause a distortion of the geometric map of the points in the graphica
l display. Nevertheless, the sensitivity of correspondence analysis to
outliers can be effectively used to verify data. Finally, based on sy
mmetry of row and column analyses correspondence analysis can be appli
ed to find out which characters can be used to construct identificatio
n keys and to selectively group variables by their importance for the
discrimination of samples.