In order to reduce the number of non occupational carbon monoxide poisoning events during winter in Beijing, the data of the daily number of CO poisoned patients from 16 emergency centers of the Beijing urban area from 2002 to 2005 was collected. The correlation between daily number of CO poisoned patients in the Beijing urban area in the winter half year (October to next April) and meteorological elements from the Beijing observatory, as well as the sea level pressure field data of areas around Beijing (8 weather stations in Hebei Province), are studied. Using the method of quasi multiple regression index probability classification, the CO poisoning index (4 classes) forecast model and the corresponding CO poisoning risk level assessment model are established. The results show that the number of CO poisoned patients varied obviously and seasonally, higher from January to March and relatively low from October to April of the next year. The historical fitting rate of completely correct and less than 1 class forecast of CO poisoning meteorological index for 24, 48 and 72 hours reached 502% to 605% and 852% to 552%, respectively. From January 2008 to April 2009, the research findings had been used in TV, newspaper, broadcast and mobile weather service.