Pseudo-feedback-based automatic query expansion yields effective retrieval performance on average, but results in performance inferior to that of using the original query for many information needs. We address an important cause of this robustness issue, namely, the query drift problem, by fusing the results retrieved in response to the original query and to its expanded form. Our approach posts performance that is significantly better than that of retrieval based only on the original query and more robust than that of retrieval using the expanded query. Categories and Subject Descriptors: H.3.3 [Information Search and Retrieval]: Retrieval Models General Terms: Algorithms, Experimentation