This article focuses on the stealthy attack design for cyber-physical systems under Kullback–Leibler divergence, where the attacker's objective is to maximize the remote error covariances while maintaining undetected. A novel historical data-based attack model with only two attack parameters is proposed. Within the framework, the attack parameters are solved analytically, which results in a better attack performance and a significant parameter reduction compared to the existing attack strategies. Finally, simulation and experiment results are given to demonstrate the proposed strategy.