This document discusses type inference on noisy RDF data. It describes how RDF data like DBpedia contains many missing type statements that limit the ability to issue structured queries. A naive approach to add types via RDFS reasoning is shown to fail on noisy data. The document then presents the SDType approach which uses statistical type predictions based on domain/range declarations and property occurrences to infer missing types. Evaluation shows SDType achieves high precision and recall on DBpedia and OpenCyc data and was used to add over 3 million type statements to DBpedia.