The multiscale coarse-graining (MS-CG) method is used to construct solvent-free CG models for polyglutamine peptides having various repeat lengths. Because the resulting CG models have fewer degrees of freedom than a corresponding all-atom simulations, they make it possible to study the self-assembly of polyglutamines at high concentrations for the first time by allowing for better equilibration and statistical sampling that is well beyond the range achievable by all-atom models. Molecular dynamics (MD) simulations performed with these models show that polyglutamine monomers with repeat lengths <= 28 fluctuate between their folded and unfolded states. Monomers with 32 or more residues are stable and form alpha-helix solid structures. The degree of monomer compactness increases with chain length in both cases. CG MD simulations of equilibrium polyglutamine aggregates show that even at high concentrations, the system statistically fluctuates between heterogeneous and homogeneous configurations, rather than simply aggregates. The degree of aggregation and fluctuation increases with concentration and chain length. All of these phenomena are consistent with the experimental observations and may be explained by a mechanism that the collective nonbonded interactions between polyglutamine molecules in water solution are only weakly attractive. Finally, this work demonstrates that computer simulation of polypeptides self-assembly and aggregation, which is presently beyond the reach of all-atom MD simulations, is attainable using solvent-free MS-CG models.