Differentiating adductor laryngeal dystonia (ADLD) and primary muscle tension dysphonia (pMTD) can be challenging. Unlike pMTD, ADLD is described as "task-dependent" with voiced phonemes purportedly provoking greater sign expression than voiceless phonemes. We evaluated the ability of two automated acoustic measures, the Cepstral Spectral Index of Dysphonia (CSID) and creak, to detect task dependency and to discriminate ADLD and pMTD. CSID, % creak, and listener ratings of dysphonia severity were obtained from audio recordings of patients with ADLD (n = 29) or pMTD (n = 33) reading two sentences loaded with either voiced or voiceless phonemes. Group × Sentence Type interaction effects confirmed that both "normalized" CSID and % creak detected task-dependent sign expression in ADLD (i.e., worse symptoms on the voiced- vs. voiceless-loaded sentence). However, a stepwise binary logistic regression analysis with group (ADLD vs. pMTD) as the dependent variable and % creak and normalized CSID variables (voiced, voiceless, and voiced vs. voiceless difference) as covariates revealed that the normalized CSID voiceless-laden sentence z score was the only significant predictor of group membership. Estimates of diagnostic precision from the normalized CSID voiceless sentence z scores were superior to % creak or listener ratings. Finally, the CSID possessed the strongest correlations with listener severity ratings regardless of group or sentence type. Although both normalized CSID and % creak detected task-dependent performance as a distinguishing feature of ADLD, a CSID profile wherein (a) the voiceless sentence z score was less severe than the voiced sentence and (b) the normalized voiceless sentence z score was within approximately 2 SDs (or less) of typical expectations provided the best estimates of diagnostic precision. Automated acoustic measures such as the CSID and creak provide useful information to objectively discriminate ADLD and pMTD.