Cognitive control is often conceptualized as a set of abilities allowing to pursue current goals. Popular models have therefore partitioned cognitive control into different subcomponents, but recent findings question the validity of some of these constructs. Prominently, the construct validity of inhibition has been questioned given the low correlations observed between performance in tasks putatively measuring it, such as the Simon and the Stroop. However, this debate is further complicated by the possibility that the observed low correlations arise due to measurement error, as congruency effects typically display low reliabilities. In the present study, we took a theoretically-informed approach to these issues. Based on classic findings in experimental research, we argue that control abilities are adequately captured only in those subsets of trials where they are most needed. Therefore, restricting data analysis to those subsets, should improve both validity and reliability of control measures. In Experiment 1 we show that focusing on trials where inhibition is most required (i.e., post-congruent and fast-response trials), significantly increased both the split-half reliability and between-task correlations of congruency effects in the Stroop and Simon tasks. In Experiment 2 we extended our theoretical approach to task switching, where we reasoned that shifting would be better captured when excluding the first trial of a repetition run. Although the switch cost is known to be a relatively reliable measure, we could show that both split-half and test-retest reliabilities where improved when excluding first repetitions. Our results highlight the utility of integrating experimental-research findings in inter-individual differences studies.