Various estimators of variance components for the balanced three-stage nested random effects model are compared under standard assumptions of normality and independence of random effects, using MSE as the measure of performance. Several results are proven, demonstrating the inequality relationships on MSE of the unbiased, restricted maximum likelihood and maximum likelihood, and several modifications of the unbiased and maximum likelihood estimators as introduced here, including Stein-type estimators. Some Bayesian estimators are also considered for numerical comparison of the MSE. Computations are carried out to assess the order of difference in the MSE's. A large class of equivariant estimators of the error component is shown to be inadmissible.