Here we describe the Semantic Pictionary Project—a set of online games and tools designed to collect large amounts of structured data about the object characteristics and perceptual properties of word referents. The project hinges on the use of encoding-decoding games and a set of creation tools to capture data using online crowdsourcing. We describe the architecture of the basic tools behind the games, the structure of the resulting data, and how this information may be integrated into existing statistical semantic models. We also describe two validations using data collected from one of the tools (2D Geon Pictionary) demonstrating typicality effects in the metrics of raw Geon objects created by subjects, and unique variance in the predictions of word pair metrics over currently used linguistic and property data.