The intra-subject variability (ISV) in brain responses during cognitive processing across experimental trials has been recognised as an important facet of neural functionality reflecting an intrinsic neurophysiological characteristic of the brain. In recent decades, ISV in behaviour has been found to be significantly associated with cognitive functioning varying across individuals, development, ages, and pathological conditions. Event-related potentials (ERPs) measured in single trials are important tools for characterizing ISV at the neural level. However, due to the overlapping spectra of noise and signals, the retrieval of information from single-trial ERPs related to cognitive processing has been a challenge. We review the major problems that researchers face in the estimation of ISV in single-trial ERPs. Then, we present an extensive evaluation of several methods of single-trial latency estimation based on both simulated and real data. The relationships of ISV in ERPs and reaction times are compared between the different single-trial methods to assess their relative efficiency in predicting task performance from neural signals. The pros and cons of the methods are discussed.