In tradition studies on prediction accuracy of GM (1, 1) model, only consider fitting error, without considering impact from accuracy of original data. Thus it is not comprehensive or scientific. Prediction error results from two sides: 1. error from modeling fitting which is irrelative to data accuracy; 2. accumulating error from data transformation which is relative to data accuracy. In this paper, by using error transfer and error synthesis theories, and considered the characteristics of grey number, we established GM (1, 1) prediction accuracy model, and then we demonstrated scientific and effectiveness of accuracy model, in the end, an example to prove that this model given.