This work proposes a criterion to select a subset of indirect measurements avoiding redundant information. The key idea of the proposal is to reduce the actual number of measurements to be performed to those strictly providing relevant information to the testing procedure, therefore keeping the incurred misclassification costs at acceptable levels. The proposed method selects the measurements presenting minimum mean and spread of the correlations between all the possible pairs formed by the measurements within the target subset. In order to show the viability of the proposal, the method has been applied to select a subset of indirect measurements in several analog filters affected by parametric variations. The obtained misclassification levels using the subset of measurements indicated by the proposed method are considerably low. In average, the number of misclassified circuits using the selected subset are within the first 2.4% of the misclassification population obtained when all the possible subsets are explored, therefore validating the proposal.