Transient evoked otoacoustic emissions of a large population of neonates (466 ears) are analyzed, with the aim of establishing if a significant latency difference can be observed between "pass" (333) and "fail" (133) ears, discriminated with a screening protocol based on band and global reproducibility. The ears that did not pass the test in at least one frequency band are named "fail," for simplicity, but they should be more appropriately defined as "partial pass." In a previous study, significantly different average latencies had been observed in adult subjects, comparing normal hearing and hearing impaired ears [J. Acoust. Soc. Am. 111, 297-308 (2002)], but no similar study has been conducted on neonates yet. An improved wavelet technique was applied to transient evoked otoacoustic emission data, to get accurate experimental measurements of the otoacoustic emission latency. The results show statistically significant differences between the latency distributions of the "pass" and "fail" populations, with the "fail" latencies longer. However, non-Gaussianity of the distributions and systematic errors associated with low reproducibility levels suggested using a conservative approach in the data analysis and interpretation. New otoacoustic estimates of cochlear tuning in neonates are also provided.