The objective of this research is to compare the one-step-ahead prediction methodsfor first-order autoregressive model with outliers. These methods are the predictionmethod using recursive mean OLS method (RM), using recursive median OLS method(RMD), and using improved recursive median OLS method (IRMD). The sample sizesare 25, 50, 100, and 250. The variance of random error (σa2) is equal to 1. Thepercentages of outliers are 5% and 10% and the magnitude of outliers is equal 3σato 5σa and. This research uses the Monte Carlo simulation method. The experimentwas repeated 10,000 times for each condition to calculate the prediction mean squareerror (PMSE). Results of the research are as follows: a RM method provides thelowest PMSE when ρ is not close to 1 in all situations. The PMSE of a RMD methodis lower than that of a RM method when ρ is close to 1 in almost all cases. A IRMD provides the lowest PMSE when ρ is close to 1 in all situations.