There are many challenges associated with service selection in a multi-user, dynamic environment. One of these challenges is the frequently changing workload. Traditional approaches to service selection, however, don't adapt to the frequently changing workload in the open and multi-user environment and can't deal with the varied workload in real time. This study is to address the problem, First, a load level based multidimensional QoS model (LLBMQoS) for services is presented; Next, a load aware dynamic service selection model (LADSSM) is proposed based on LLBMQoS to optimize service selection in dynamic environments. The model adopts a two-phase service selection framework, which generates candidate queues in design time and dynamically selects the services in terms of the current workload in execution time. Finally, simulation experiment results are provided to show the proposed model can adapt to the varied workload in a multi-user environment and provide an optimal service selection scheme while meeting the end-to-end QoS constraints.