Abstract: Time series discords are subsequences that are maximally different to all other time series subsequences of a longer time series. Discord Detection is widely used in time series applications. We observed that the discord position are often changed when noise data interfere with the time series. This phenomenon is produced because the traditional method cannot concern the factor that noise data infect the original data’s distribution. In this study, we opposed a novel method which combined top-k discord detection with uncertain ranking to achieve uncertain top-k discord detection. Through transforming the discord score interval to satisfied with Gaussian distribution, the new method can ranking series data with arbitrary distribution. Finally, we demonstrate a comprehensive experimental study to verify the effectiveness and efficiency of the proposed approach.