Understanding the internal states of others is essential in social exchanges. The aim of the presented thesis is to provide a deeper understanding of the impact nonverbal cues have on the perception of internal states, namely of emotions and associated cognitive appraisals. First, we explored naturalistic behaviour from a hidden camera, described with technical coding systems, FACS for perceived facial muscle movements and a coding scheme defined by ourselves for the hand, arm and torso movements. Participants were asked to judge observed persons' internal states. These descriptions of behaviours and perceptive judgments allow us to make a link between internal state attributions and concrete physical expressions. Second, a novel method was used for expression exploration by transposing naturalistic behaviours to a virtual agent, Greta, which enables a fine tuning of expressions. In order to improve the synchronisation of behaviours the Multimodal Sequential Expression model was created for the Greta agent. Complex expressions were manipulated, one cue at a time, and expression were judged by participants, who were asked to attribute internal states to the agent. Results support the componential approaches to expression, in which particular cues are considered meaningful.