Abstract: To achieve modeling and identification for systems with valve stiction nonlinearities, unlike existing methods that mostly rely on numerical optimization and assume known time delays, a model-reconstruction-based joint identification method is proposed for valve stiction Hammerstein systems with an unknown time delay. First, an enhanced Li stiction model is developed and derived in a discrete recursive form, and then embedded into an Autoregressive moving average with exogenous input model to construct a parameter-separated stiction Hammerstein identification model. Second, an extended model is formed by augmenting the parameter vector based on a redundant-rule scheme, so as to eliminate the effect of the unknown delay on the model structure. Finally, data filtering and hierarchical identification strategies are introduced, and a filtered rolling-window hierarchical forgetting-gradient recursive subalgorithm is designed to estimate the model parameters; the time delay is determined via a discrete search and redundant parameters are removed. Simulation and experimental results demonstrate good estimation accuracy and predictive performance, validating the effectiveness of the proposed method.