Inferring a user's high-level goals from low-level sensor readings has been drawing increasing attention from both AI and Pervasive Computing communities recently. A common assumption made by most approaches is that a user has a single goal in mind or aims to achieve several goals sequentially. However, in real-world environments, a user often has multiple goals concurrently carried out and a single action can serve as a step towards multiple goals. In this thesis, we formulate the sensor-based multiple-goal recognition problem and exemplify it in an indoor environment where an RF-based wireless network is available. We propose a recognition algorithm to infer a user's multiple high-level goals from low-level sensory data. In our approach, we establish a model set where goal recognition models are instantiated and terminated dynamically. Each model is a finite state machine and evolves over time among pre-defined states to perform recognition. By distinguishing the state of a model. we can infer whether one of a user's goal is present or not. Experiments with real data demonstrated that our method can accurately and efficiently recognize multiple goals in a user's trace. Thus, we provide a general framework for goal recognition and achieve a major advance over previous work.