This paper demonstrates a framework of processes for identifying potential witnesses of events from evidence they post to social media. The research defines original evidence models for micro-blog content sources, the relative uncertainty of different evidence types, and models for testing evidence by combination. Methods to filter and extract evidence using automated and semi-automated means are demonstrated using a Twitter case study event. Further, an implementation to test extracted evidence using Dempster Shafer Theory of Evidence are presented. The results indicate that the inclusion of evidence from micro-blog text and linked image content can increase the number of micro-bloggers identified at events, in comparison to the number of micro-bloggers identified from geotags alone. Additionally, the number of micro-bloggers that can be tested for evidence corroboration or conflict, is increased by incorporating evidence identified in their posting history.