Like electroencephalographic (EEG) activity during reaction time tasks, EEG activity during tasks without overt responses may also consist of two components: stimulus-locked and -unlocked components. The extraction of such stimulus-unlocked components has been difficult owing to the unknown delays. Here, we propose a novel method to extract both of the two components from single-channel EEG epochs. In this method, we initially set random values for the delays and extract uncontaminated stimulus-locked and -unlocked components using the preset delays and a discrete Fourier transform. Then, we reconstruct the EEG by overlapping the extracted components with the preset delays, and calculate the residual errors between the reconstructed and original EEG. This procedure is repeated by updating the delays until the residual errors become adequately small. After verifying the performance of this method by two kinds of simulations with artificial and EEG data, we apply the method to EEG during NoGo trials of a Go/NoGo task, and obtain the stimulus-unlocked components, the magnitudes of which are comparable with those of the stimulus-locked components. By applying this method, it is possible to study internal and subjective brain activity, which occurs with variable delays.