Patient‐derived xenograft (PDX) models are widely used as preclinical cancer models and are considered better than cell culture models in recapitulating the histological features, molecular characteristics and intratumoral heterogeneity (ITH) of human tumors. While the PDX model is commonly accepted for use in drug discovery and other translational studies, a growing body of evidence has suggested its limitations. Recently, the fidelity of cancer cells within a PDX has been questioned, which may impede the future application of these models. In this review, we will focus the variable phenotypes of xenograft tumors and the genomic instability and molecular inconsistency of PDX tumors after serial transplantation. Next, we will discuss the underlying mechanism of ITH and its clinical relevance. Stochastic selection bias in the sampling process and/or deterministic clonal dynamics due to murine selective pressure may have detrimental effects on the results of personalized medicine and drug screening studies. In addition, we aim to identify a possible solution for the issue of fidelity in current PDX models and to discuss emerging next‐generation preclinical models.