Molecular dynamics (MD) simulation is an important approach for conformational search. The conformational searching efficiency in MD simulation is greatly limited by the systematic conformational change or motion. We developed a new MD simulation method to enhance the conformational search efficiency through accelerating the systematic motion. In this work, we describe the theoretical basis and the simulation algorithm of this method for atomic systems. In this method, systematic motion is accelerated by a guiding force derived from a local free-energy surface defined for a system. Under certain approximations, the guiding forces can be estimated from the force information the system experienced in the past through a memory function, and these forces are used to guide the current motion in the same simulation. Therefore, this guiding force is called the self-guiding force and this kind of simulation is called the self-guided molecular dynamics (SGMD) simulation. We have performed detailed analysis of the characteristics of the SGMD in terms of energetic, structural, and dynamic properties with a Lennard-Jones argon system. In addition, we demonstrated the enhanced conformational search efficiency of the SGMD method through the ergodic measure and the crystallization of liquid argon.