BACKGROUND: Previous research indicates that Thompson and colleagues' (Thompson, 2001; Thompson and Shapiro, 2005) Treatment of Underlying Forms (TUF) can efficiently remediate agrammatic sentence-processing deficits. The theoretical basis of TUF is that training production of complex, noncanonical sentence structures can concomitantly improve production of untrained, syntactically related but simpler sentence structures. Whereas this generalisation to untrained syntactic forms has been well established within constrained, sentence-level tasks, which exploit the same response modality used during training, TUF's generalisation potential in terms of cross-modal effects and discourse-level improvements requires further exploration. AIMS: DM, a 52-year-old male with an agrammatic Broca's aphasia profile, was provided with a modified version of TUF, which targeted his writing skills and included a Discourse Training Module that allowed direct rehearsal of targeted syntactic frames within a discourse context. The hypotheses tested were as follows: (a) DM would improve his written production of trained sentence structures and demonstrate generalisation to untrained exemplars of targeted sentence structures as well as untrained, syntactically related syntactic structures; (b) written sentence production treatment would facilitate gains in DM's spoken production of trained and related, untrained sentence structures; and (c) DM would exhibit improved sentence production abilities in discourse post-treatment. METHODS and PROCEDURES: A single subject, multiple baseline across-behaviours design was implemented to evaluate acquisition of trained sentence types (object- and subject-extracted embedded who-question sentences), to discern generalisation to untrained sentence types (object- and subject-extracted matrix questions, passives) and discourse, or both, and to identify maintenance of treatment effects. Each week, DM completed two 90-minute sessions of modified TUF as well as written sentence production homework. OUTCOMES and RESULTS: DM displayed a pattern of sentence acquisition typical of TUF recipients, generalising gains in complex sentence production to the production of untrained, less complex, theoretically related structures. Gains in written production generalised to spoken production of the same structures, and improvements across predominately pragmatic versus morphosyntactic discourse variables were also noted. CONCLUSIONS: The treatment outcomes of a modified, written version of TUF were comparable to those in previous studies (e.g., Ballard and Thompson, 1999), and indicated that training written sentence production can evoke substantial cross-modal generalisation to speech. Despite inclusion of a Discourse Training Module, pragmatic versus morphosyntactic aspects of DM's discourse showed most improvement. Therefore, continued investigation of TUF is recommended to determine whether it can efficiently treat structural aspects of discourse production, or what modifications will ensure generalisation to discourse contexts in a broader spectrum of aphasic patients.