Abstract: In the mixed linear model with two variance components, we propose a class of the estimators of the variance component. The class of estimators includes many popular estimators, such as ANVOA estimator. In this class we discuss the necessary conditions of admissibility. Using the necessary conditions we get two new estimators which are better than ANOVA estimator. At the same time we discuss the nonnegative estimators of variance components and suggest two new estimators which are better than ANOVA estimator.