Inhibition of beta-Site amyloid precursor protein cleaving enzyme 1 (BACE1) has been shown to be an effective treatment for Alzheimer's disease. A wealth of research has focused on finding highly selective small-molecule inhibitors targeting the BACE1 over its close homologue BACE2 to avoid potential side effects. However, given the highly structural similarities of BACE1 and BACE2, designing highly selective BACE1 inhibitors remains a huge challenge. Recently, it has been reported that a potential BACE1 inhibitor named C28 (similar to 52-fold selectivity) exhibited greater selectivity to BACE1 over BACE2 than the previously reported inhibitors AZD3293 and AZD3839 (similar to 1.5-fold and 14-fold selectivity). However, few computational studies have been performed to reveal its underlying mechanisms. In this study, a series of molecular modeling techniques were performed to reveal the selective mechanisms. Classical molecular dynamics (cMD) simulations indicated that the major variations appeared to be controlled by overall protein dynamics. Free energy calculations further suggested that the binding affinities of AZD3293 to BACE1 and BACE2 are similar, but the binding affinity of AZD3839 and C28 to BACE1 is much higher than to BACE2, and that the major variations are electrostatic interactions. The protein dynamics and energy differences were further observed in accelerated molecular dynamics (aMD) simulations. In addition, the umbrella sampling simulations revealed the inhibitors' different patterns of dissociation from the binding pockets of BACE1 and BACE2, and that different energy barriers were responsible for the selectivity. The physical principles revealed by this study may facilitate the rational design of more potent BACE1 selective inhibitors. Communicated by Ramaswamy H. Sarma