Marine Abnormal Event (MAE) is an abnormal decrease or increase of marine environmental parameters, which covers a specified spatial domain and lasts for a specified temporal duration. Marine abnormal event can provide the temporal and spatial characteristics of regional sea-air interactions and global climate change, which has an important scientific significance. Based on the above information, we propose a novel algorithm to extract the MAE from the long-term raster datasets, named as MAESTEM (Marine Abnormal Event Spatio-Temporal Extraction Method). The MAESTEM has three key steps: the extraction of MAE at a temporal dimension, the extraction of MAE at a spatial dimension, and the tracking of MAE. At the temporal dimension, each grid pixel within an image is taken to be the one-dimensional time series, and its mean and standard deviation are taken as the criteria to define its abnormal snapshot status. If and only if the abnormal snapshot and its subsequent ones are not smaller than the specified threshold, i.e. T, the abnormal snapshots are defined as a temporal MAE, denoted by TMAE. We utilize the spatial neighborhood statistics method to count the number of spatial neighborhoods of a raster pixel which belongs to TMAE and to obtain the marine abnormal events at the spatial dimension, denoted as SAME, by using the spatial dimension abnormal extracting method. In the final step, we use the spatial topological relationship of SMAE to identify whether the SAMEs at the previous and post snapshots belongs to the same event. If they overlap, they are considered being in the same event. If the temporal duration of the event exceeds the temporal threshold, save the event, otherwise, delete it. Finally, the Pacific Ocean is taken as a research area, and its monthly averaged sea level anomaly (SLA), which is obtained from the remote sensing imagery during a period from January 1993 to December 2012, is used to test the feasibility and efficiency of MAESTEM.