We discuss problems and features of current semiempirical molecular orbital techniques and test some of the approximations and assumptions used. Prerequisites for a ''next generation'' technique include orthogonalization corrections, effective core potentials and an implicit dispersion term. However, validation of experimental parameterization data using density functional theory or the Gaussian 2 approach reveals significant errors in some cases. Developers of future methods will need to validate all their parameterization data and may no longer be able to parameterize for heats of formation at 298 K, but may need to use Born-Oppenheimer binding energies. We also suggest that there is no inherent reason that semiempirical techniques should not reproduce hydrogen bonding and show that the Gaussian potentials added to the core-core terms in AM1 and the PMn methods actually weaken hydrogen bonding, rather than strengthening it.