This study examined whether there are different subtypes of developmen
tal dyslexia. The subjects were 51 dyslexic children (reading below th
e 30th percentile in isolated word recognition), 51 age-matched normal
readers, and 27 younger normal readers who scored in the same range a
s the dyslexics on word recognition. Using methods developed by Castle
s and Coltheart (1993), we identified two subgroups who fit the profil
es commonly termed ''surface'' and ''phonological'' dyslexia. Surface
subjects were relatively poorer in reading exception words compared to
nonwords; phonological dyslexics showed the opposite pattern. However
, most dyslexics were impaired on reading both exception words and non
words compared to same-aged normal readers. Whereas the surface dyslex
ics' performance was very similar to that of younger normal readers, t
he phonological dyslexics' was not. The two dyslexic groups also exhib
ited a double dissociation on two validation tasks: surface subjects w
ere impaired on a task involving orthographic knowledge but not one in
volving phonology; phonological dyslexics showed the opposite pattern.
The data support the conclusion that there are at least two subtypes
of developmental dyslexia. Although these patterns have been taken as
evidence for the dual-route model, we provide an alternative account o
f them within the Seidenberg and McClelland (1989) connectionist model
. The connectionist model accounts for why dyslexics tend to be impair
ed on both exception words and nonwords; it also suggests that the sub
types may arise from multiple underlying deficits. We conclude that pe
rformance on exception words and nonwords is not sufficient to identif
y the basis of dyslexic behavior; rather, information about children's
performance on other tasks, their remediation experiences, and the co
mputational mechanisms that give rise to impairments must be taken int
o account as well.