In many important application domains such as Business and Finance, Process Monitoring, and Security, huge and quickly increasing volumes of complex data are collected and archived. Strong efforts are underway developing automatic and interactive analysis tools for mining useful information from these data repositories. At the heart of many data analysis tools is the notion of similarity which has to be defined appropriately in the given data space to allow meaningful clustering, classification, and retrieval, among other tasks. The feature vector approach is one of the most popular schemes for managing multimedia data. For many data types such as audio, images, or 3D models, an abundance of different feature vector extractors are available. The automatic (unsupervised) identification of the best suited feature extractor for a given multimedia database is a difficult and largely unsolved problem. We address the problem of comparative unsupervised feature space analysis. We propose two approaches for the visual analysis of certain feature space characteristics contributing to estimated discrimination power provided in the respective feature spaces. We apply the approaches on a database of 3D objects represented in different feature spaces, and we experimentally show the methods to be useful (a) for unsupervised comparative estimation of discrimination power and (b) for visually analyzing important properties of the components (dimensions) of the respective feature spaces. The results of the analysis are useful for feature selection and engineering. The second approach, projection-based data space similarity visualization, is interesting (a) for visual discrimination analysis within a given similarity space, and (b) for comparative analysis of similarity characteristics of a given data set represented in different similarity spaces. We introduce an intuitive and effective novel approach for projection-based similarity visualization for interactive discrimination analysis, data exploration, and visual evaluation of metric space effectiveness. The approach is based on the hull metaphor for visually aggregating sets of points in projected space and can be used with a variety of different projection techniques. The effectiveness of the approach is demonstrated by application on two well-known data sets. We also present statistical evidence justifying the hull metaphor. We advocate the hull-based approach over the standard symbol-based approach to projection visualization, as it allows a more effective perception of similarity relationships and class distribution characteristics. This work carried out by Tobias Schreck and Christian Panse in a DFG-supported project following the V3D2 Strategic Research Initiative (Project of D. Keim and D. Saupe).