As the number of web services with the similar function is increasing, QoS-based service selection at runtime has become an important research topic. The existing QoS-based services selection approaches always assume that the QoS data coming from service providers and users are effective and trustworthy, which is actually impossible in reality. This paper proposes a service selection approach considering the trustworthiness of QoS data, which classifies and computes the QoS attributes according to the source of QoS data. For the QoS attributes whose data come from service providers, the statistics of past runtime data to revise the providers' QoS data are used. For the QoS attributes whose data come from users, feedback similarity to weigh users' QoS data is used. Furthermore, an implementation framework and a set of experiments are given, which show that this approach can effectively weaken the influence of untrustworthy QoS data on the services selection, thus can strengthen the accuracy of the service selection.