In this extended abstract, we provide a brief synopsis of our published, ongoing and future work on data driven exploration of perceptual properties of 3D shapes. We focus on research problems related to 3D data in massive online libraries which are difficult to search, organise and reuse. Our approach involves the use of crowdsourcing and machine learning methods to mitigate the challenges. We collect a considerable amount of human judgements on the style matching and aesthetics of 3D shapes by leveraging crowdsourcing platforms. Our method demonstrates that machine learning and crowdsourcing are suitable techniques to build data-driven models of visual perceptual attributes of 3D shapes. We build data-driven models of two perceptual characteristics, namely style similarity and aesthetics. As demonstrated by our results, the learned models can be employed in various 3D graphics applications such as search, organisation, scene composition, and visualisation of 3D shape data.