The prediction accuracy of the burnup characteristics of large liquid-metal fast breeder reactors (LMFBRs) is very hard to evaluate because of the unavailability of the direct experimental information. A quantitative evaluation was performed on the accuracy of the burnup property by use of the sensitivity coefficients in a large LMFBR and the covariance matrix of nuclear data. Also evaluated was the decrease in prediction error when the cross-section set was adjusted by the use of experimental data, such as criticality, reaction rate ratios, and others. It was concluded that accuracy with the direct use of current nuclear data is ±30% for burnup reactivity loss and ±5% for breeding ratio. On the other hand, the accuracy would be improved to ±18% and ±2.5% by utilizing the experimental data obtained on the zero-power plutonium reactor assembly.