We have constructed a response adaptive clinical trial to treat patients sequentially in order to maximize the total survival time of all patients. Typically the response adaptive design is based on the urn models or on sequential estimation procedures, but we used a bandit process in this dissertation. The objective of a bandit process is to optimize a measure of sequential selections from several treatments. Each treatment consist of a sequence of conditionally independent and identically distributed random variables, and some of these treatment have unknown distribution functions. For the purpose of this clinical trial, we are focusing on the bandit process with delayed response. These responses are lifetime variables which may be censored upon their observations. Following the Bayesian approach and dynamic programming technique, we formulated a controlled stochastic dynamic model. In addition, we used an example to illustrate the possible application of the main results as well as "R" to implement a model simulation.