Assessment of the maximum tolerated dose for a small sample of patients is the objective of phase I trials in oncology. We propose a new adaptive approach performing differentiation of each type of toxicity and their grades, definition of the dose-limiting toxicity as a logical combination of toxicity events, modeling of the risk of toxic events as a function of the drug dose, and integration of individual covariates. An application study with retrospective data from a phase I Taxol pediatric trial revealed age as an influential covariate, allowing individualization of a patient's maximum tolerated dose. A simulation study illustrates the practice of sequential inclusion of patients, of constraints in the dose-finding process and of trial stopping rules. The approach presented here allows quick maximum tolerated dose assessment with a small sample of patients and assists the design of phase I trials in oncology.