A Sigma-Mu approach is proposed for mutual funds portfolio selection. The mean and variance of the overall performance of each asset are considered, according to an additive aggregation model, subject to weights’ preferences provided by the decision maker. These preferences concern two independent sets of weights, i.e., those pertaining to the investment indicators and those pertaining to the time periods associated with the estimation of the indicators. For the first time in the Sigma-Mu framework, a weighting matrix is exploited, assisting on the development of a method to appraise the sources of variance, due to the weighting scheme of either the indicators or the periods. The Mu's, Sigma's and covariances estimated according to the Sigma-Mu approach, enter as inputs to mixed-integer quadratic programming (MIQP) mean-variance portfolio optimization models, in order to implement an empirical testing procedure, for a period of 8 years. The underlying MIQP models are equipped to consider non-convex investment policy constraints, such as the number of securities to be included in the portfolio, specific binary buy-in thresholds, the desired exposure of the portfolio to each investment advisor etc. The dataset that has been chosen for the empirical testing includes European mutual funds, that offer a broad exposure to the whole span of investment strategies and styles. The results document that the suggested approach may effectively be utilized in mutual funds investment management, since the portfolios constructed by the suggested methodology are associated with superior absolute and risk-adjusted performance against benchmarks.