The paper introduces the Deep Potential Molecular Dynamics (DeePMD) method, a scalable and accurate scheme for molecular simulations. DeePMD uses a deep neural network (DNN) to model interatomic forces and potential energy, trained with *ab initio* data. The DNN preserves natural symmetries and is "first principle-based," avoiding *ad hoc* components. DeePMD provides efficient and accurate results for various systems, including bulk materials and molecules, with results indistinguishable from *ab initio* data at a computational cost that scales linearly with system size. The method overcomes limitations of traditional force fields by capturing the complex, nonlinear dependence of atomic energies on local environments. DeePMD is demonstrated to be effective in reproducing trajectories of extended and finite systems, showing high accuracy in energy, forces, and virial predictions. The method's scalability and efficiency make it a promising tool for large-scale molecular simulations, addressing the long-standing dilemma of balancing accuracy and efficiency in molecular dynamics.The paper introduces the Deep Potential Molecular Dynamics (DeePMD) method, a scalable and accurate scheme for molecular simulations. DeePMD uses a deep neural network (DNN) to model interatomic forces and potential energy, trained with *ab initio* data. The DNN preserves natural symmetries and is "first principle-based," avoiding *ad hoc* components. DeePMD provides efficient and accurate results for various systems, including bulk materials and molecules, with results indistinguishable from *ab initio* data at a computational cost that scales linearly with system size. The method overcomes limitations of traditional force fields by capturing the complex, nonlinear dependence of atomic energies on local environments. DeePMD is demonstrated to be effective in reproducing trajectories of extended and finite systems, showing high accuracy in energy, forces, and virial predictions. The method's scalability and efficiency make it a promising tool for large-scale molecular simulations, addressing the long-standing dilemma of balancing accuracy and efficiency in molecular dynamics.