Despite significant progress, we show that state of the art 3D human pose andshape estimation methods remain sensitive to partial occlusion and can producedramatically wrong predictions although much of the body is observable. Toaddress this, we introduce a soft attention mechanism, called the PartAttention REgressor (PARE), that learns to predict body-part-guided attentionmasks. We observe that state-of-the-art methods rely on global featurerepresentations, making them sensitive to even small occlusions. In contrast,PARE's part-guided attention mechanism overcomes these issues by exploitinginformation about the visibility of individual body parts while leveraginginformation from neighboring body-parts to predict occluded parts. We showqualitatively that PARE learns sensible attention masks, and quantitativeevaluation confirms that PARE achieves more accurate and robust reconstructionresults than existing approaches on both occlusion-specific and standardbenchmarks. The code and data are available for research purposes at {\small\url{https://pare.is.tue.mpg.de/}}