Preclinical models often fail to capture the diverse heterogeneity of human malignancies and as such lack clinical predictive power. Patient-derived tumor xenografts (PDX) have emerged as a powerful technology: capable of retaining the molecular heterogeneity of their originating sample. However, heterogeneity within a tumor is governed by both cell-autonomous (e.g., genetic and epigenetic heterogeneity) and non-cell-autonomous (e.g., stromal heterogeneity) drivers. Although PDXs can largely recapitulate the polygenomic architecture of human tumors, they do not fully account for heterogeneity in the tumor microenvironment. Hence, these models have substantial utility in basic and translational research in cancer biology; however, study of stromal or immune drivers of malignant progression may be limited. Similarly, PDX models offer the ability to conduct patient-specific in vivo and ex vivo drug screens, but stromal contributions to treatment responses may be under-represented. This review discusses the sources and consequences of intratumor heterogeneity and how these are recapitulated in the PDX model. Limitations of the current generation of PDXs are discussed and strategies to improve several aspects of the model with respect to preserving heterogeneity are proposed.