The purpose of this paper is to present dynamic approach to selection of efficient port folios using a forecasts of variances and covar iances from the multivaria te GARCH models. Evaluation of efficiency for different methods of asset allocation is also performed for stocks from the WSE. Twelve specif ications o f the multivariate GARCH models, the univariate GARCH model and six other covariance matrix estimation methods are used. Taking in to consideration time varying variances and covariances of stock returns in portfolio selections increases, with some exceptions , efficiency of asset allocation process. Simple specifications of the multivariateGARCH models, which parame ters are estimated in one stage, are the best performing models. From economic point of view, the differences between the models are not significant, with exception of the factor and orthogonal models. RiskMetrics methodology commonly used by practitioners does not give good results for constructions of efficient portfolios .