We investigate the impact of reset attacks in cyber-physical systems. Two types of attack models, basic and advanced reset attacks, are considered in this paper. For the basic attack, which only injects a constant to the state estimate without knowing the system parameters, we provide a sufficient and necessary condition under which the adversary cannot destabilize the systems. On the other hand, advanced reset attacks, which can inject a predesigned sequence with full knowledge of system parameters, can be more devastating to cyber-physical systems. Depending on the goals of the adversary, we further categorize advanced reset attacks into Type-I advanced reset attack, where the adversary aims to drive the system estimation error to infinity, and Type-II advanced reset attack, where the goal of the adversary is to drive the system state to some other target states. A sufficient and necessary condition is presented for the existence of Type-I advanced reset attack. A closed-form optimal Type-II advanced reset attack is obtained by dynamic programing. Simulation results are provided to illustrate the effectiveness of both advanced reset attacks.