Network theory has been used to study interaction systems at largely different scales and many measures have been developed to analyze the network representation of such systems. These analyses highlighted strikingly recurring features, suggesting general underlying principles. With the aim of investigating the possible role of robustness as one of these principles, we developed a zero-parameter theory, which we termed Buffered Qualitative Stability (BQS). BQS is able to predict many features to be expected in complex interaction networks that need to be robust despite a high level of noise. These predictions are consistently verified in experimentally validated genetic networks of different organisms, thus supporting the role of BQS as a general tool to characterize and understand robustness in interaction systems.