The context of this work is the search for realism and believability of Virtual Humans. Our contribution to achieve this goal is to enable Virtual Humans (VH) to react to spontaneous events in virtual environments (VE). In order to reflect the individuality of each VH, these reactions have to be expressive and unique. In this paper we present firstly a model of reaction based on personality traits. The model was defined using statistical analysis of real people reacting to unexpected events. We also consider that the emotional state is involved in the modulation of reactions, thus we integrate a model of emotion update. Secondly, we present a semantic-based methodology to compose reactive animation sequences using inverse kinematics (IK) and key frame (KF) interpolation animation techniques. Finally, we present an application that demonstrates how Virtual Humans can produce diferent movements as reaction to unexpected stimuli, depending on their personality traits and emotional state.