It is necessary to determine seismic wave first breaks before the following processing of initial field seismic data is done. These first breaks separate seismic signals into two parts: microseismic noise and a useful information part. In the paper the authors present the overview information about the existing approaches aimed at automatic first breaks detection. All the described first break picking approaches are separated into two types: the one that uses neural networks  and the other one that composes the classic approaches without using neural networks. Meanwhile, neural network-based approaches could include the classic ones in themselves. It was found out that nowadays the neural network-based approaches prevail with the dominating quantity of scientific publications on the topic. The authors classify the classic approaches into the threshold algorithm, the $STA/LTA$ fractal dimension estimation algorithm, the higher-order statistics calculation algorithm, the autoregressive algorithm, the filter picker algorithm, the correlative approach, the dynamic time warping algorithm and the fuzzy clusterisation algorithm. The neural network approaches include fully connected dense networks, Kohonen networks and convolutional networks. The brief description of each approach is given in the paper, providing references for a reader to be able to get more information if needed. In conclusion, the authors provide a general scheme summarising considered automatic first break picking approaches. The accuracy table achieved by those methods is also provided.