Modeling systems that are governed by van der Waals (dispersion) interactions using empirically corrected DFT methods is becoming increasingly popular due to the promise of a CCSD(T) level accuracy at the computational cost of DFT. Although, DFT methods are computationally efficient in comparison to the CCSD(T) method, currently, structural optimizations using DFT methods are generally only feasible for systems of less than a few hundred atoms. We seek a method applicable to macromolecular complexes. In order to model such large systems, empirically corrected semiempirical methods appear to be an attractive alternative. As with most common DFT methods, the popular semiempirical methods (e.g., AM1) also do not model long-range dispersion (and therefore an empirical correction term is desirable), but this is not their only shortcoming. For weakly interacting systems, hydrogen bonding also poses a concern. A new empirically corrected AM1 method that uses two empirical correction terms, one for dispersion and one for hydrogen bonding interactions, is presented and termed AM1-FS1. This new empirically corrected AM1 method has been parametrized to a diverse training set of 66 complexes that includes nonequilibrium structures and yields sub-kilocalorie accuracy in the prediction of intermolecular interaction energies. More significantly, AM1-FS1 achieves this result with substantially less parametrization than existing empirically corrected semiempirical methods and without modification of the original AM1 parameters so that it retains both the computational efficiency and predictive power for thermo-chemical quantities of the original AM1 Hamiltonian. The performance of AM1-FS1 is also tested on several carbon nanostructure complexes and pseudorotaxanes and is found to produce results in very good agreement with the best first-principles calculations.