Query-based biclustering can be used to explore public gene expression data for genes coexpressed with genes of interest to a certain researcher (the query). These methods, however, fail when faced with a list of query-genes with diverse expression profiles. In addition, a threshold on the minimal coexpression with the query-genes needs to be defined in advance. To deal with these problems we introduce an ensemble approach for query-based biclustering. The method relies on a specifically designed consensus matrix in which the biclustering outcomes for multiple query-genes and for different possible coexpression thresholds are merged in a statistically robust way. Graph clustering is used to obtain non-redundant consensus biclusters from this matrix. We tested out different ensemble construction schemes and illustrate the effectiveness of this approach