Abstract: The localization of multiple odor sources using a mobile robot in a natural airflow environment is studied. Firstly, the olfactory perception model is constructed, and the odor concentration and airflow direction/speed detected by the robot in each time step are fused as the evidence about the existence of odor sources in local area. Then, the evidence is combined with the existing evidences by using the evidence theory to update the spatial distribution of the multiple odor sources. The experimental results in outdoor natural environment show that the proposed olfactory perception model is applicable to the online mapping of multiple odor sources in environment with time-varying airflow, and the proposed mapping method based on the evidence theory can achieve a better robustness compared with the IP (independence of posteriors) algorithm (a Bayesian occupancy grid mapping method).