Aqueous arginine solutions are used extensively for inhibiting protein aggregation. There are several theories proposed to explain the effect of arginine on protein stability, but the exact mechanism is still not clear. To understand the mechanism of protein cosolvent interaction, the intraprotein, protein−solvent, and intrasolvent interactions have to be understood. Molecular dynamics simulations of aqueous arginine solutions were carried out for experimentally accessible concentrations and temperature ranges to study the structure of the solution and its energetic properties and obtain insight into the mechanism by which arginine inhibits protein aggregation. Simulations of proteins (α-chymotrypsinogen A and melittin) were performed. Structurally, the most striking feature of the aqueous arginine solutions is the self-association of arginine molecules. Arginine shows a marked tendency to form clusters with head to tail hydrogen bonding. Due to the presence of the three charged groups, there are several possible configurations in which arginine molecules interact. At relatively high concentrations, these arginine clusters associate with other clusters and monomeric arginine molecules to form large clusters. The hydrogen bonds between arginine molecules were found to be stronger than those between arginine and water, which makes the process of self-association enthalpically favorable. From the simulation of the proteins in aqueous arginine solution, arginine is found to interact with the aromatic and charged side chains of surface residues. A probable mechanism of the effect of arginine on protein stability consistent with our findings is proposed. In particular, arginine interacts with aromatic and charged residues due to cation−π interaction and salt-bridge formation, respectively, to stabilize the partially unfolded intermediates. The self-interaction of arginine leads to the formation of clusters which, due to their size, crowd out the protein−protein interaction. The mechanisms proposed in the literature are analyzed on the basis of the simulation results reported in this paper and recent experimental data.