The recent development of Virtual Reality (VR) technologies and interactive visual content faces some important challenges in multimedia processing and Quality of Experience (QoE). Considering omnidirectional, also called 360-degree, content , one major research topic is the development of reliable visual attention models. In this paper, we present a new dataset to study the influence of emotions on eye behavior in omnidirec-tional content. This dataset is based on an eye-tracking experiment where 19 observers have assessed emotional valence and arousal dimensions in 360-degree images. Several analyses are then conducted to compare fixation and saccade features, inter-observer saliency congruency and spatial oculomotor biases in positive, neutral and negative content. Results show a significant impact of negative images on visual attention, with more visual agitation and avoidance behavior from larger, longer and faster saccades. However, no obvious difference of eye behavior is found between positive and neutral stimulis on this dataset.