Cameras for eye tracking are becoming standard for modern virtual reality head-mounted displays. While evaluation and interaction of two-dimensional eye tracking data are widely researched and applied, the three-dimensional point of regard is mostly unexplored. Usually, gaze depth is determined using the intersection of a gaze ray with a two-dimensional plane such as the screen. However, this requires complete scene knowledge and an occlusion-free environment. In this work, the estimation is done by intersecting the two eyes' gaze rays. This enables the usage of semi- transparent objects a user can interact with by focusing directly on them or looking through them. Insight into the users' gaze depth could provide valuable data to help understand user behavior, disambiguate targets in high occlusion environments and enable new interaction techniques. The subject of this work is to evaluate the estimation of gaze depth and to explore possible novel use cases. In order to get a stable estimation of the gaze depth, two vergence-based calibration procedures are developed improving on existing research and comparing modalities that could impact calibration quality. The implementations were evaluated in a small-scale pilot (n=10) study. Results show that interaction works well in distances up to 1.2 meters, while very close distances of about 30 cm were perceived as uncomfortable to look at by some users. Results also indicate that a continuously moving calibration target leads to a better all-purpose calibration. Carefully calibrating the space used for interaction can therefore lead to improved gaze depth estimation and interaction.