Some transition metal borides are ultra hard. While not harder than diamond, they are easier to process and can be cheaper, sparking intense interest. However, we so far cannot predict which particular borides should be ultra hard. A striking example is the three structurally similar diborides of Ti, Re, and Os, among which only ReB 2 is ultra hard. For this trio, using a combination of theory and experiment done on both the solids and small cluster models, we show that the nature of the metal-boron bonds is the key to hardness, in contrast to the existing theory, which overlooks metal-boron interactions. Ti-B bonding is purely ionic in TiB 2 , and the material yields to shear stress like graphite. OsB 2 is highly covalent, with both bonding and antibonding Os-B backbonds present, which weaken the Bnetwork, and ease the OsB 2 yield to compression. ReB 2 has only the bonding Re-B σ-backbond, which strengthens the material against both shear and compression. A general strategy for ultra hard boride design is proposed.