We propose a non-parametric estimator of the autoregression parameter in the first-order stationary autoregressive (AR) process in which the responses are subject to right censoring. Simulation studies based on Monte Carlo training samples are used to evaluate, compare and contrast the performance of the estimator with that of a parametric estimator for censored AR models that is available in the literature. The simulation results show that the performance of the new estimator is comparable with that of the parametric estimator. However, the new estimator has an advantage over the parametric estimator because it does not require knowledge of the error distribution and hence, unlike the parametric estimator, can be applied to a wide variety of situations.
