Abstract: ‍ ‍Brain-computer interface （BCI） is a technology that connects the brain with external devices to achieve information exchange. The core is to bring the advantages of the human brain into play and interact with the external environment directly through the human brain. It has become a research hotspot all over the world. P300 is a typical component of event-related potentials triggered by a specific stimulus paradigm. Initial studies found that when the human brain is stimulated by certain small probability events， a positive peak with a latency of about 300 ms appears in the EEG signal. Because the P300 component has stimulation specificity， it has been widely used to study cognitive functions such as attention and recognition. The EEG paradigm based on the P300 feature has also become one of the three mainstream paradigms. In recent years， there have been a large number of applications with practical functions combining P300 and brain-computer interface， and related decoding algorithms have been proposed and continuously improved， which makes it convenient for people to study P300 brain-computer interface. However， the related decoding algorithms of the P300 brain-computer interface are not systematically summarized， and the characteristics of the algorithms have not been clearly compared. In this paper， a P300 EEG decoding algorithm based on non-spatial information is proposed according to the feature that P300 components mainly respond in the parietal-occipital region. Based on the different response characteristics of P300 components in different brain regions， a P300 EEG decoding algorithm based on spatial information is proposed. Based on the rapid development of information technology in recent years and the characteristics of deep learning in the application of P300 brain-computer interface， a new P300 EEG decoding algorithm based on deep learning is proposed. Starting from the above three perspectives， this paper summarizes the research progress of P300 brain-computer interface decoding algorithms in recent years and the characteristics of deep learning in the application of P300 brain-computer interface， a new P300 EEG decoding algorithm based on deep learning is proposed. Starting from the above three perspectives， summarize the research progress of P300 brain-computer interface decoding algorithms in recent years， and compare the performance of algorithms based on existing P300 public datasets， clarify the advantages， disadvantages， and application scope of each algorithm. Finally， further explore the existing typical problems and potential research directions in the future.