The pelves of the child dummies of the widely used Hybrid-III family are based on minimal data from children. Because an accurate pelvis design is critical for realistic assessments of belt restraint interactions, an analysis of medical imaging data was conducted to develop guidance for improved pelvis design. Three-dimensional polygonal meshes of the bony pelvis were generated from computed tomography (CT) data from 81 children from ages 5 to 11. After aligning to a uniform anatomical coordinate system, the meshes were resampled to create a quadrilateral mesh with 12,960 vertices for each pelvis. A principal components analysis was conducted with the mesh vertex coordinates and the locations of 31 landmarks. Over 99% of the variance in size and shape was accounted for by the first 40 components. A three-dimensional model representing the target for a new dummy pelvis was developed using bispinous breadth as the predictor variable. To obtain the appropriate geometry for the six-year-old (6YO) and 10YO Hybrid-III dummies, a regression analysis was conducted using a large sample of child anthropometry data from a previous study to obtain a target dimension for bispinous breadth, using the design stature for each dummy as input. A separate regression analysis was conducted to predict principal component scores from bispinous breadth. Reconstructing a pelvis model from the principal components scores predicted for the target bispinous breadth values yielded a three-dimensional design target for the each dummy. The new pelvis target is similar in overall size to the current pelvis components, but the location of the anterior-superior iliac spine is markedly lower, which has important implications for belt interaction.