Verbal working memory resources may impact syntax comprehension. Thirteen Italian children with cochlear implants (CIs) were assessed in relative clause (RC) comprehension, digit span and nonword repetition and compared to 13 chronological age peers (CA) and 13 younger controls (LA) with normal hearing (NH). The RC comprehension task tested subject relatives (SR), object relatives with preverbal (OR) and postverbal subjects (ORp) where number features were manipulated. Children with CIs show worse performance than controls in RC comprehension and nonword repetition. In the RC task, number features facilitated comprehension by children with NH, but not by children with CIs. The memory measure that predicted RC comprehension was digit span. In the LA group, backward digit span predicted comprehension of all RC types. Forward digit span predicted comprehension of ORs with number mismatch in the CA group, and comprehension of ORs with number mismatch and ORps in children with CIs. In these conditions, high memory resources are needed to exploit number features in theta-role assignment.