We present a pipelined architecture for a text simplification system and describe our implementation of the three stages- analysis, transformation and regeneration. Our architecture allows each component to be developed and eval-uated independently. We lay particular emphasis on the discourse level aspects of syntactic simplification as these are crucial to the process and have not been dealt with by previous research in the field. These aspects include generating referring expressions, deciding determiners, deciding sentence order and preserving rhetorical and anaphoric structure.