New media are increasingly used to capture ambient geographic information in multiple contexts, from mapping the evolution of the Tahrir Square protests in Egypt to predicting influenza outbreaks. The social media platform Twitter is popular for these applications; it boasts over 500 million messages (“tweets”) generated every day from as many total users at an average rate of 5,700 messages per second. In the United States, Twitter is used to communicate prescribed agricultural or other burning and the emergence, response to, and containment of wildfires. A prototype for operational prescribed and wildland fire detections from social media, specifically Twitter, was created using natural language processing and information retrieval techniques. The intent is to identify and locate prescribed burns and wildland fires in the contiguous United States often missed by satellite detections with the hope of providing relevant, spatio-temporal fire data for emission estimates, inventories and burned area mapping efforts. The authors present their methodology and an evaluation of its performance in collecting relevant tweets, extracting metrics such as area burned and geo-locating the fire events using the GeoNames geographic gazette. Compared to two operational satellite fire products, this data mining effort mapped fires potentially unknown to the satellite record.