This chapter discusses the detection of abnormal events. Events are special activities or behaviors. Abnormal events are those events that are significantly different from normal and regular activities or behaviors. From the perspective of time and space, abnormal events often do not occur frequently over time and may have certain degree of locality in space. From a statistical perspective, the probability of its occurrence should be relatively small. The main problem encountered in abnormal event detection is that the its probability is quite small, and there are fewer samples available. Therefore, there are many methods that use normal samples for training, while considering the difference between the current situation and the normal situation during detection to determine the abnormality. This chapter will first discuss the classification of abnormal event detection methods, taking into account the features used and the learning approach. This chapter will present several detection networks, such as based on convolutional auto-encoders, based on sparse auto-encoder, based on LSTM auto-encoder. This chapter will introduce two depth generative models: one considers both generative loss and temporal smoothing loss, other is scene content adapted. This chapter will also present a master-auxiliary aggregation strategy.