Remaining useful life (RUL) estimation is an integral part of prognostics and health management of engineering systems. Using RUL estimation, failures could be detected and maintained before they occur. Advancements in sensor technology permitted monitoring the health state of engineering systems producing a lot of data. This data is utilized for building data-driven estimation models. Today, both in practice and literature data-driven models for RUL estimation are very popular since the construction of these models usually does not require domain knowledge, it is easy to adapt existing models for different engineering systems and provide accurate estimations. Nonlinear Auto-regressive neural network with eXogenous inputs (NARX) model is used for dynamic modeling and multi-step long-term forecasting. NARX makes predictions by iteratively simulating the dynamic behavior of underlying nonlinear time series. Since the system monitoring data are dynamic nonlinear time series, this feature of NARX makes it a suitable model for RUL estimation. In this study, a genetic algorithm-based parallel mode NARX (GABPM-NARX) model is proposed for RUL estimation. The proposed approach is applied to NASA C-MAPSS dataset from the NASA data repository. The results of the analysis are satisfactory showing that the proposed GABPM-NARX model is a promising method for RUL estimation.