This study investigated the acoustic realization of disyllabic Low tone (T3T3) words in Mandarin. A large corpus of 850 hours of journalistic speech was used. The segmentation of the continuous Mandarin speech was carried out using the LISN (former LIMSI) speech transcription system in forced alignment mode. The full corpus was aligned twice using two different pronunciation dictionaries, one with and one without systematic tone variants. We analyzed the tonal representations of the underlying T3T3 sequences that the LISN system produced with the two alignment strategies. We further investigated the effect of different factors, namely prosodic position, word frequency, tonal contexts, and parts of speech, on conditioning the surface realization of the underlying T3T3 sequences. The LISN outputs suggest that the first T3 of disyllabic T3T3 words is not always realized as T2 in continuous speech and a range of factors conditions its acoustic realizations.