This document discusses outsourcing FrameNet annotation to crowdsourcing. It presents a two-step and simplified one-step methodology for crowdsourcing frame and semantic role annotation. Experiments using these methods on the CrowdFlower platform showed that the simplified one-step approach had higher accuracy and was faster than the two-step approach. Lessons learned include that definitions need to be simplified for non-experts and negation and modality are difficult concepts. Further research directions include larger-scale experiments and linking entities to structured knowledge bases.