A Bayesian model-based framework for phase I dose escalation studies is presented. The advantages of such approaches are discussed in the context of phase I cancer trials. Critical aspects of the statistical model and prior specification are presented, along with the need to reflect the uncertainty of estimated rates of toxicity, which allows for patient risk monitoring (overdose control). Other components of the study design such as inference and decision making are discussed and the importance of both good statistical guidance and clinical expertise for on-study decisions is highlighted. Examples detailing the model building and calibration are presented for a trial, along with the model recommendations in first few cohorts. A range of possible prior specifications (non-informative, animal data, historical data from humans, mixture priors) are briefly discussed as well.