Last year, in the context of the INEX evaluation initiative, we could show that our retrieval system IRStream is successfully applicable as a retrieval engine for XML-documents. Nevertheless, we have to point out that IRStream can be
further optimized in many directions.

In the present paper we show, how IRStream was extended and improved for its application to INEX 2003 in order to achieve better retrieval results. Furthermore, we present some first retrieval results, which demonstrate the impact of the improvements of IRStream concerning the quality of
the retrieval result.