Videos may induce users' mixture emotions. Most present emotional tagging research ignore the phenomena of multiple emotions' coexistence and mutual exclusion. In this paper, we propose a novel emotional tagging approach by exploring multiple emotion's relations. First, several visual and audio features are extracted from videos. Second, support vector machines are used as the classifiers to get the measurements of emotional tags. Then, a Bayesian network is adopted to learn the relationships among emotional tags. After that, the Bayesian network is used to infer the video tags combining the measurements obtained by support vector machines. Experiments on a dataset of 72 affective videos demonstrate the effectiveness of our approach.