Obtaining spontaneous emotional expressions is the very first and vital step in affective computing studies, for both psychologists and computer scientists. However, it is quite challenging to record them in real life, especially when certain modalities are required (e.g. 3D representation of the body). Traditional elicitation and capturing protocols either introduce the awareness of the recording, which may impair the naturalness of the behaviors, or cause too much information loss. In this paper, we present natural emotion elicitation and recording experiments, which were set in child-robot interaction scenarios. Several state-of-the-art technologies were employed to acquire the multi-modal expressive data that will be further used for emotion modeling and recognition studies. The obtained recordings exhibit the expected emotional expressions.