This paper investigated the performance of an optimal bandpass filtering technique to effectively identify click evoked otoacoustic emissions in neonates. The frequency response of the filter was optimized by maximizing the correlation coefficient between two replicate recordings (reproducibility). The optimal cut-off frequencies were also set by controlling the maximum energy loss after filtering to stop the crucial response contents being filtered along with the noise. As an additional constraint, the cut-off frequencies were forced to be outside the range of 1.6 to 4.2 kHz to avoid (in a response identification task) the spontaneous otoacoustic emissions from polarizing the cut-offs and thereby filtering fundamental frequency components, thus leading to an erroneous response classification. The best performance of the optimal filter in terms of increasing the post-filtering reproducibility was obtained when the procedure was applied to recordings whose reproducibility before filtering ranged between 60 and 80 per cent, i.e. for responses classified in most cases as partial pass.