The classical picture of a functioning cell is that the proteins and RNAs are doing all of the work while the DNA containing all of the instructions sits in splendid isolation in the nucleus. In this point of view, any DNA which does participate in the making of proteins or RNAs would be considered junk, because it does not contribute in an obvious way to the cell's activities. In fact, quite a lot of mammalian non-coding DNA (about a third of the human genome) strongly resembles a parasitic overgrowth of retrovirus-derived seqeuences known as repetitive elements, like kudzu overgrowing the back of a barn.
Yet population genetics analysis of this hinterland frequently show it being defended over time against mutations, suggesting that it must contribute to fitness. Thus it has to be contributing somehow to the survival of the organism. A recent report in Nature (Nature, subscription) confirms this by comparing two closely related species of the fruit fly Drosophila.
The findings of this study to challenge the picture that junk is really junk. The way out suggested in the Nature minireview seems to be that mammalian genomes might be junky but smaller organisms like flies much less so. For example, flies seem to have fewer repetitive elements than humans, but the mouse genome seems human-like. The main point of both the work and the opinion is that given the large mass of non-coding DNA, even a statistically small role could really matter in the long run.