Abstract: The objective of this article was to determine the effects of several characteristics of data structure on variance of prediction error using both sire and animal models. Data were collected according to mixed model in Ceylanp?nar State Farm of Dairy Cattle`s Breeding Unit. Data were included milking records from 1987 to 1999. Data structures were replicated 300 times for each combination of variance and covariance assumption and proportion of occupied subclasses. Data were also evaluated in mixed models appropriate on sire and animal models. Results were comprised correlations between variables and variance of prediction error was obtained by evaluating the model and scheme. Simple curvilinear regression analysis was used to study several design variables in further details and to determine the effects on variance of prediction error. According to results, the sire models yielded a wider range of values for the design variables and in animal model analysis, the number of animals with a progeny test, individual test, and their combination, were 0.34, 0.41 and 0.25, respectively. Progeny tests yielded larger variances of prediction error than did individual performance. Genetic connections were not strongly associated with the variance of prediction error. No single piece of information was useful for predicting accuracy and several other contributions to accuracy are necessary.