In Cloud computing, users with different service requirements often need to negotiate with service provider via Service Level Agreement (SLA). The unique pay-as-you-go billing way in Cloud computing challenges resource provisioning for service providers. In this paper, based on the Dirichlet multinomial model, we present an efficient reputation-based QoS provisioning scheme, which can minimize the cost of computing resources, while satisfying the desired QoS metrics. Unlike the previous counterparts, we consider the statistical probability of the response time as a practical metric rather than the typical mean response time. Numerical results show the efficiency and effectiveness of the proposed scheme.