Recently Wang, Gould, and Stwalley [Phys. Rev. Lett. 80, 476 (1998)] showed by exciting atomic fragments that predissociation occurs in the 1(g) and 0(u)(+) states of K-2. In this study, we supplement the interpretation given in the original report with a more detailed discussion. The calculations presented here use discrete variable representation, close coupling, and semiclassical methods, and incorporate an ab initio calculation of diagonal and off-diagonal spin-orbit elements. The predicted predissociation widths for the 1(g) state are too narrow to be observed directly, but may be correlated with the strength of the spectral lines observed by exciting atomic fragments. The missing v=90 level of the 1(g) state in the observed fragmentation spectrum, where the predissociation rate is calculated to be nearly maximum, is attributed to a node in the absorption line strength close to this vibrational level. Since the available information on K-2 electronic states (despite recent progress) gives inadequate information on the inner walls of the A state and especially of the b state, calculations of predissociation in the 0(u)(+) state can give only a range of possible widths. We fit widths to 0(u)(+) spectral lines observed by monitoring trap loss. The fitted experimental widths are significantly larger than the maximum calculated predissociation widths, suggesting that there was line broadening associated with the laser power required for the observation.