Theoretical models and computational techniques are useful for gaining insight into the interactions, movements, and functions of atoms and molecules, ranging from small chemical systems with few atoms to large biological molecules with many atoms. Due to the inability of force field methods to accurately describe different properties of metalloenzymes and the prohibitive computing cost of high-level quantum methods, computationally efficient models are needed. This dissertation describes the development of new quantum semiempirical models for metalloproteins. The original AM1 (Austin Model 1) based on the neglect of diatomic differential overlap approximations was re-parameterized to describe the structural and energetic properties of biomolecules that mimic the active sites of metalloproteins. The biologically inspired genetic algorithm PIKAIA was used to optimize the parameters for each chemical element. Structures and energies of various clusters analogous to complexes found in metalloproteins were prepared as a training set using hybrid density functional theory. Models were trained to reproduce all of the properties included in the small training set. The optimized models were validated for large testing sets that incorporate bigger complexes and related reactions. Finally, the optimized models were used to study biologically-relevant processes in condensed phase using molecular dynamics simulations. All the gas- and liquid-phase results from the optimized models were compared with original semiempirical models as well as available high-level theoretical and experimental results. Metal ions play crucial roles in biological systems. They actively participate in structural, catalytic, and co-catalytic activities of a large number of enzymes. The development of semiempirical models is divided into three parts. First, new AM1 parameters for hydrogen and oxygen were developed to describe gas-phase proton transfer reactions in water and static and dynamic properties of liquid water. Gas-phase results were compared with original AM1, RM1, and PM3 models, whereas liquid results were compared with original AM1, AM1-W, and AM1PG-W models, and with available experimental results. It is found that the optimized model reproduces experimental data better than other available semiempirical models. Second, using the previously optimized model for hydrogen and oxygen, the AM1 model is re-parameterized for zinc and sulfur to describe important physical and chemical properties of zinc, water, hydrogen sulfide complexes mimicking structural motifs found in zinc enzymes. Metal-induced pKa shifts are computed for water and hydrogen sulfide, and compared with available theoretical and experimental results. Third, using previously optimized parameters for hydrogen, oxygen, and zinc, AM1 parameters for carbon and nitrogen are optimized to study proton transfer, nucleophilic attacks, and peptide hydrolysis mechanisms in zinc metalloproteases. Overall, the optimized models give promising results for the various properties of biomolecules in gas-phase clusters and in condensed phase. Particularly, the water model reproduces the proton transfer related properties in gas-phase and the structure, dielectric properties, and infrared spectra of liquid water. The zinc/sulfur model reproduces the hydration structure of zinc cation and zinc-bound hydrogen sulfide. Results for the coordination configurations of zinc solvated in water and in hydrogen sulfide confirm the versatility of the model. The optimized model for carbon and nitrogen improves the overall performance compared to AM1 and PM3. The optimized model for carbon and nitrogen reproduces structures and various energetic terms for zinc-ligands systems (representing the active sites of zinc enzymes) when compared to density functional theory results. The optimized model can be used to study metal-ligand reactivity in zinc enzymes.