In this paper technique8 are presented which allow the evaluation of accuracy of prediction methoda, with special reference to the prediction of gradand yield. The more prominent methods of predicting graeeland yield are diecussed. Two coefficients, which are normalizatione of the mean-aquared-error, are propoaed 8s measure8 of accnracy. It ia shown that the man-squarederror partitiom into the w m of three meaningful componenta of error. The diatribution of the componenta is diecussed under the aaanmption that the actual and predicted yielde have a bivariate normal diatribution. The problem of accuracy analysis is addreeeed when one is only interested in ranking the yields of graesland 8re.U. The ratio of the mean-squared-error for two competing prediction methods ie inveetigated. Finally, a multiplicative model ie studied which allow8 predicting ability to differ over time or over region& Key words: Accuracy analysis, relative accuracy, Theil's h p o s i t i o n of error, ped&%on of graashnd yield, t a t s of h y p o t h k , Speurman~s rank correkztion, Kendall's r coefficient, mdtiplicative dhxnnpsitiOn.