The assignment of collections belonging to the genus Trema has been reasses
sed using numerical taxonomy methods. Observation in the field in Togo, Wes
t Africa, confirmed high phenotypic variation but not the previously assert
ed lack of character concordance within the T. guineensis complex. This stu
dy was carried out to assess the circumscription of taxa in the area. Dista
nce between specimens was computed using the Cower's general coefficient of
similarity and the non-parametric MODECLUS cluster analysis was used to di
scover hew the specimens segregated. The existence of three clusters in the
158 specimen dataset using 44 morphological characters was suggested by MO
DECLUS. Canonical discriminant analysis supported the recognition of those
three clusters using 40 morphological characters. Classificatory discrimina
nt analysis showed that all specimens except one are identified correctly b
y the discriminant function. We suggest that three species be recognized in
Togo and the neighboring countries: T. orientalis, T. africana, and T. nit
ens. A key to the species is included and the advantages of the method sugg
ested here are discussed.