Abstract: Hadoop is one of the most popular cloud computing platforms．In this platform,the existing access control model adopts Kerberos for identity verification,combines with authorization mechanism based on ACL,and uses the Delegation Token and Block Access Token,realizing a simple access control mechanism．There is an obvious shortco-ming in this model,namely,it considers only the identity authenticity of a user while authorizing,nevertheless the credibi-lity of its following behaviors．Once access control right is granted,there won’t be any kind of supervision．This paper proposed a new trust-based access control model in Hadoop,which is based on the existing access control model in Hadoop and is called LT．LT sets a trust value for each user,updates this value according to users’ behavior records,and controls the user to access Hadoop cluster with the trust value dynamically．Comparing with the existing access control model in Hadoop,the access and authorization mechanism realized in LT isn’t a one-time access and authorization,but a thoroughly real-time and dynamic process,so LT is more secure,more flexible and has a finer control particle size．Experiments show that this model is not only right and effective but also overcomes the disadvantage on lacking of security about the existing access control model in Hadoop.It can control a user to access or use the resources supplied by a Hadoop cluster dynamically and effectively.