Our participation in TREC 2004 aims to extend and improve the use of the DFR (Divergence From Randomness) models with Query Expansion (QE) for the robust track. We experiment with a new parameter-free version of Rocchio's Query Expansion, and use the information theory based function, InfoDFR to predict the AP (Average Precision) of queries. We also study how the use of an external collection affects the retrieval-performance.