Urnov et al. are trying to effect gene therapy - where a mutation causing a genetic form of severe combined immunideficiency (SCID) (also known as the bubble boy syndrome). Affected SCID patients can have little to no immunity to infection what so ever. SCID in this case is caused by a single site mutation in the IL2R-gamma gene.
Their method is to use a DNA binding protein (containing zinc finger domains that bind the DNA) which attracts a DNA repair enzyme.
The result is that 18% of the cells have the SCID gene repaired. This would have been a trivial result if they had applied selection, which generally refers to any method which eliminates cells which were not affected (in this case having their DNA repaired) by a transformation like their treatment. They claim that 18% of all the cells were repaired (I assume this is in a cell culture). This implies that if their treatment were applied to living tissue, 18% of the cells would be repaired in situ, which should be enough to restore immune function.
While its possible to apply a selection method to gene therapy, it would be far better if you did not. Cancer chemotherapy for instance is a selection based treatment for instance, relying on toxins which kill fast dividing or growing cells preferentially, leaving the slower growing tissues relatively viable. Still its quite toxic and disruptive to the patient.