The treatment of cancer is continually improving as a result of a better understanding of oncogenesis and the development of new targeted compounds. Early clinical trials evaluating such candidate compounds require a large number of patients, and are expensive, time consuming, and expose patients to certain risks. To select the most effective molecules, preclinical investigation of antitumor compounds is an important step in the drug development process. Three main categories of preclinical cancer models are generally used in preclinical investigations, namely genetically engineered models, xenografts derived from human tumor cell lines, and human tumor fragments from patients implanted directly into immunodeficient mice, known as tumorgrafts. The establishment of tumorgrafts constitutes a long-term process consisting of various steps, in which the final objective is to show that the validated model accurately reproduces human cancer, with a high predictive value of therapeutic efficacy (regardless of the type of treatment), and closely mimics clinical situations frequently observed in patients with cancer, such as resistance to standard treatments, metastases, and relapse after initial therapies (involving residual tumor-initiating cells). The aim of this study is therefore to discuss the proposed criteria for the establishment and validation of preclinical models of human cancers that should be used for further pharmacological assessments.