This paper considers the optimum replacement times of a steam generator in nuclear power plant with failure data. It is assumed that the failure pattern of units is given as a Weibull distribution and repair and periodic preventive maintenance are performed periodically. The maximum likelihood method is used to estimated the Weibull parameters of failure distribution from failure data. Relpacement, output-decresing and maintenance costs are considered to determine the optimal replacement times by simulation. Numerical examples are included with actual failure data and cost estimators.