This study examined the postulate that training production of syntactically complex sentences results in generalization to less complex sentences that have processes in common with treated structures. Three agrammatic aphasic patients were trained to produce wh-movement structures, object clefts and/or object extracted who-questions, while generalization between these structures was tested. One NP-movement structure, passive sentences, also was tested for control purposes. Wh-movement occurs from the direct object position to specifier position in the complementizer phrase [SPEC, CP] for both wh-movement structures. In who-questions movement occurs in the matrix sentence, whereas, in object clefts movement occurs within an embedded relative clause, rendering them the most complex. Results showed robust generalization effects from object clefts to matrix who-question for 1 participant (D.L.); however, no generalization was noted from who-questions to object clefts for another (F.P.), and 1 participant (C.H.) showed acquisition of who-questions, but not object clefts, during the baseline condition without direct treatment. As expected, none of the participants showed improved production of passives. These findings supported those derived from our previous studies, indicating that generalization is enhanced not only when target structures are related along dimensions articulated by linguistic theory, but also when the direction of treatment is from more to less complex structures. The present findings also support proposals that projections of higher levels in the syntactic treatment are dependent on successful projection of lower levels. For our participants, training movement within CP in a lower (embedded) clause resulted in their ability to project to CP at higher levels. (JINS, 1998, 4, 661–674.)