Conspicuous differences between ab initio harmonic frequencies for the nu(6) and nu(17) hydrogen bridge stretching modes of diborane and those calculated from an empirical harmonic force field have remained unexplained for nearly 20 years. This work conclusively demonstrates that an unexpectedly large anharmonicity effect is responsible for the discrepancy. Cubic and semidiagonal quartic force constants obtained with coupled-cluster theory and a reasonably large basis set are used to calculate fundamental frequencies. For the bridge BH stretching (BHb) modes, all theoretical values are within 15 cm(-1) of the actual fundamental frequencies, while corresponding harmonic frequencies are well above the empirical values. It thus appears that BHb-stretching anharmonicity is considerably greater than previously supposed, being roughly three times that (as measured by the magnitude of 1 - nu/omega) for terminal BH bonds. This is likely to be a general feature of electron-deficient compounds.