The applicability of the Zung self-rating depression scale (SDS) in pregnancy is unknown. We aimed to identify redundant items and evaluate the Zung SDS's structural validity. Two samples of pregnant women were invited from two districts in Shanghai (Yangpu sample, n = 6468 and Huangpu sample, n = 402). The Yangpu sample was randomly split into YGroup1/2/3. Item's properties were evaluated via the item response theory in YGroup1. Exploratory and confirmatory factor analyses were correspondingly executed in YGroup2 and YGroup3. Those items with discrimination parameter (α) lower than 0.65 or factor loading smaller than 0.4 were deleted from the scale. The final structure was validated in the Huangpu sample. Items 4 (sleep), 7 (weight loss), 8 (constipation) and 9 (tachyarrhythmia) exhibited low discrimination power. Items 2 (diurnal variation), 5 (appetite), 10 (fatigue) and 19 (suicide idea) made a low contribution to all factors. A three-factor model was eventually constructed as cognitive (Items 14, 16, 17, 18 and 20), psychomotor (Items 6, 11 and 12) and affective (Items 1, 3, 13 and 15). The Zung SDS needs modification before applied to pregnant women in China. The items describing the overlap symptoms of the physical change in pregnancy and mood disorder should be deleted.