The output of reverse-engineering methods for biological networks is often not a single network prediction, but an ensemble of networks that are consistent with the experimentally measured data. In this paper, we consider the problem of combining the information contained within such an ensemble in order to (1) make more accurate network predictions and (2) estimate the reliability of these predictions. We review existing methods, discuss their limitations, and point out possible research directions toward more advanced methods for this purpose. The potential of considering ensembles of networks, rather than individual inferred networks, is demonstrated by showing how an ensemble voting method achieved winning performance on the Five-Gene Network Challenge of the second DREAM conference (Dialogue on Reverse Engineering Assessments and Methods 2007, New York, NY).