Neuronal response onset latency provides important data on the information processing within the central nervous system. In order to enhance the quality of the onset latency estimation, we have developed a ‘double sliding-window’ technique, which combines the advantages of mathematical methods with the reliability of standard statistical processes. This method is based on repetitive series of statistical probes between two virtual time windows. The layout of the significance curve reveals the starting points of changes in neuronal activity in the form of break-points between linear segments. A second-order difference function is applied to determine the position of maximum slope change, which corresponds to the onset of the response. In comparison with Poisson spike-train analysis, the cumulative sum technique and the method of Falzett et al., this ‘double sliding-window, technique seems to be a more accurate automated procedure to calculate the response onset latency of a broad range of neuronal response characteristics.