Xenotransplantation of human cancers into immunodeficient mice is a very useful approach for studying human tumor biology. However, the occasional occurrence of lymphomagenesis in some mice can spoil the model and must be investigated in detail. We found that a high percentage (32.5%, 26/80) of cancer patient-derived xenografts (PDXs) resembled lymphoma in NOD/SCID mice. Of the 26 xenografts, 23 were human-derived expressing human CD45 (hCD45+) and proved to be of the B-cell subtype (CD3-/CD20+), and they were all positive for Epstein-Barr virus (EBV). The remaining 3 xenografts proved to be mouse-derived for both hCD45- and negative amplification of a human gene. The most interesting finding is that gastric cancer had much higher rates (24/126, 19.0%) of lymphoma formation in the PDX model than did colorectal cancer (1/43, 2.3%). Statistical analysis revealed that cancer type and inflammation in the parent tumor are significantly associated with lymphomagenesis. Further validation discovered lymphomagenesis by inoculating only gastritis mucosa. Therefore, our findings suggest that it is necessary to take precautions when directly xenografting cancer tissues with remarkable baseline inflammation, such as gastric cancer into immunodeficient NOD/SCID strains. Further, the established xenograft models should be validated by both leukocyte markers and human gene signatures.