This paper discusses an evaluation of an augmented reality (AR) multimodal interface that uses combined speech and paddle gestures for interaction with virtual objects in the real world. We briefly describe our AR multimodal interface architecture and multimodal fusion strategies that are based on the combination of time-based and domain semantics. Then, we present the results from a user study comparing using multimodal input to using gesture input alone. The results show that a combination of speech and paddle gestures improves the efficiency of user interaction. Finally, we describe some design recommendations for developing other multimodal AR interfaces.