The usefulness of the event-related potential (ERP) method can be compromised by violations of the underlying assumptions, for example, confounding variations of latency and amplitude of ERP components within and between conditions. Here we show how the ERP subtraction method might yield misleading information due to latency variability of ERP components. We propose a solution to this problem by correcting for latency variability using Residue Iteration Decomposition (RIDE), demonstrated with data from representative go/no-go experiments. The overlap of N2 and P3 components in go/no-go data gives rise to spurious topographical localization of the no-go–N2 component. RIDE decomposes N2 and P3 based on their latency variability. The decomposition restored the N2 topography by removing the contamination from latency-variable late components. The RIDE-derived N2 and P3 give a clearer insight about their functional relevance in the go/no-go paradigm.