Different users or the same user using controlled versus free-text vocabularies could generate different queries for the same information need. It has been known in the information retrieval literature that different query representations may retrieve different sets of documents. In this paper, we first generate multiple query vectors from a given information problem by using different relevance feedback methods. Then, we combine the multiple query vectors into a single query vector. We also show through experiments that significant improvements can be achieved by the combination of the multiple query vectors.