A novel remaining useful life(RUL)prediction method based on unscented Kalman filter(UKF)is proposed for structure with fatigue crack in machinery systems, which mainly includes two parts: performance evaluation of fatigue crack and RUL prediction. In the first part, a discrete state-space model is established based on the Paris law. Then the UKF is applied to estimate the two unknown Paris' law constants C and m combining with the real-time information obtained by sensors, in order to alleviate the negative influence on prediction accuracy caused by the uncertainty of incompletion of status information, as well as environmental noise. In the second part, the RUL of fatigue structure is predicted based on the discrete crack growth model according to the estimated result obtained by the UKF. The numerical experiments indicate that the UKF accurately identified the unknown parameters, furthermore, better performance in RUL prediction is obtained by comparing with extended Kalman filter(EKF)method. The RUL prediction accuracy can be efficiently improved by combining the discrete Paris law with UKF.