A novel approach to predict the atomic densities of amorphous materials is explored on the basis of Car-Parrinello molecular dynamics (CPMD) in density functional theory. Despite that determination of the atomic density of matter is crucial in understanding its physical properties, no such method has ever been proposed for amorphous materials until now. In our approach, by assuming that the canonical distribution of amorphous materials is Gaussian distribution, we generate multiple amorphous structures with several different volumes by CPMD simulations and average the total energies at each volume. The density is then determined to be the one that minimizes the averaged total energy. In this study, this approach is implemented for amorphous silicon ({$a$-Si}) to demonstrate its validity, and we have determined the density of {$a$-Si} to be 4.1 % lower and its bulk modulus to be 28 GPa smaller than those of the crystal, which are in good agreement with experiments. We have also confirmed that generating samples through classical molecular dynamics simulations produces a comparable result and validates our assumption. The findings suggest that the presented method is applicable to other amorphous systems, including those that lack experimental knowledge.