Randomized controlled trials conducted in resource-limited settings have shown that once women with depressed mood are evaluated by specialists and referred for treatment, lay health workers can be trained to effectively administer psychological treatments. We sought to determine the extent to which community health workers could also be trained to conduct case finding using short and ultrashort screening instruments programmed into mobile phones. Pregnant, Xhosa-speaking women were recruited independently in two cross-sectional studies (N = 1,144 and N = 361) conducted in Khayelitsha, South Africa and assessed for antenatal depression. In the smaller study, community health workers with no training in human subject research were trained to administer the Edinburgh Postnatal Depression Scale (EPDS) during the routine course of their community-based outreach. We compared the operating characteristics of four short and ultrashort versions of the EPDS with the criterion standard of probable depression, defined as an EPDS-10 ≥ 13. The prevalence of probable depression (475/1144 [42 %] and 165/361 [46 %]) was consistent across both samples. The 2-item subscale demonstrated poor internal consistency (Cronbach's α ranged from 0.55 to 0.58). All four subscales demonstrated excellent discrimination, with area under the receiver operating characteristic curve (AUC) values ranging from 0.91 to 0.99. Maximal discrimination was observed for the 7-item depressive symptoms subscale: at the conventional screening threshold of ≥10, it had 0.97 sensitivity and 0.76 specificity for detecting probable antenatal depression. The comparability of the findings across the two studies suggests that it is feasible to use community health workers to conduct case finding for antenatal depression.