This paper explores different covariance matrix estimators; both the conditional and the unconditional versions, obtained via intraday data and named realized measures, to the minimum variance portfolio selection problem. Intraday data are sampled in a synchronized manner as well as in an unsynchronized version. For sake of comparison, we also use daily data estimators. The major contribution of this work is of empirical nature, focused on the Brazilian scenario. We evaluate some out-of-sample performance indexes of the obtained portfolios for a set of 30 stocks traded on the São Paulo stock exchange (BM&FBovespa). The results show that the estimator of the conditional covariance matrix of returns using a scalar vt-VECH model based on higher frequency data leads to substantial earnings, reducing portfolio risk, increasing the average adjusted-by-risk return and decreasing the turnover.