Abstract Patient-derived xenografts (PDX) and the Avatar, a single PDX mirroring an individual patient, are emerging tools in preclinical cancer research. However, the consequences of intratumor heterogeneity for PDX modeling of biomarkers, target identification, and treatment decisions remain underexplored. In this study, we undertook serial passaging and comprehensive molecular analysis of neuroblastoma orthotopic PDXs, which revealed strong intrinsic genetic, transcriptional, and phenotypic stability for more than 2 years. The PDXs showed preserved neuroblastoma-associated gene signatures that correlated with poor clinical outcome in a large cohort of patients with neuroblastoma. Furthermore, we captured spatial intratumor heterogeneity using ten PDXs from a single high-risk patient tumor. We observed diverse growth rates, transcriptional, proteomic, and phosphoproteomic profiles. PDX-derived transcriptional profiles were associated with diverse clinical characteristics in patients with high-risk neuroblastoma. These data suggest that high-risk neuroblastoma contains elements of both temporal stability and spatial intratumor heterogeneity, the latter of which complicates clinical translation of personalized PDX–Avatar studies into preclinical cancer research. Significance: These findings underpin the complexity of PDX modeling as a means to advance translational applications against neuroblastoma. Cancer Res; 78(20); 5958–69. ©2018 AACR.