The languages of the Tamang group (Tibeto-Burman, Nepal) are presently in the process of developing tones from an earlier opposition of voicing on word-initial consonants (“tone split” in an old two-tone system: voiceless onsets > 2 high tones; voiced onsets > 2 low tones). In this group of languages, we observe a large amount of variation between dialects, between speakers, and within speakers concerning the realization of the features which define each tone. We interpret these variations as different degrees of retention of earlier features (Mazaudon, 2012). A previous phonetic study of the Tamang dialect of Risiangku (Mazaudon & Michaud, 2008) demonstrated the use of F0 and phonation, and to a lesser degree of initial stop voicing, as cues to the four tones of the system. The retention of breathiness and stop initial voicing with low pitch tones may be explained by mutual enhancement (Silverman, 1997). In the Tamang dialect of Taglung, one of the two low tones has evolved into a high falling tone. However, we still observe occasional retention of word-initial voicing (word-medial voicing is not contrastive), not only on the present-day low tone, but also on the historical low tone which is no longer low. The conditioning of this retention is the object of the present study.