<p>Alarm systems are important for industrial process safety, but they often suffer from a large number of nuisance alarms. Nuisance alarms not only submerge real alarms but also decrease operators’ reliance on alarm systems. This paper proposes a novel method to identify chattering alarms as one type of common nuisance alarms. The main idea of the proposed method is to determine whether the distribution of alarm durations follows a geometric mixture distribution. As the key contribution, the alarm durations of chattering alarms following a geometric mixture distribution are concluded based on Gaussian mixture distribution theory by supposing that the noise contained in process variables is independent and identically distributed. The proposed method has more extensive application scenarios than existing methods owing to its advantages, being capable of identifying chattering alarms from alarm data sequences containing real alarms and being unrelated to the states of process variables. Numerical and industrial cases are provided to illustrate the effectiveness of the proposed method.</p>