Static corrections are an essential step in land seismic data processing. The most preferable static correction technique is aligning the traces according to first arrival waves in refracted signals. There have been many automatic first arrival algorithms introduced and applied in the industry since the seismic data size grows. All the automatic methods need post processing of the results to eliminate the mis-picks. Therefore, manual detection of the first arrival waves is still the most reliable method together with automatic and semi-automatic methods. With increasing number of data channel in seismic surveys, manual picking is turned into an expensive process. Therefore there is a need of an automatic first arrival detection method which produces trustworthy results and also allow an easy elimination of the mis-picks if exists. In this thesis, an automatic first break detection method, Support Vector Machines (SVMs) classification, is introduced and tested together with four conventional methods. The advantage of the introduced method is the availability to use the results of any other first break picking method and algorithm as a feature. It is based on the detection of similar first arrival waveforms and therefore any similarity measure can also be used as an input in the purpose of first break detection. Moreover, the result of the classifications can be used for post processing step to remove the mis-picks. Three type of seismic dataset are chosen to test the algorithms to compare the efficiency: A noise-free synthetic, a 2D vibroseis and a 3D with a dynamite source. The detection accuracy of the SVMs classification is more than 98 % for the all tested datasets whereas the maximum accuracy of the conventional methods are 81 %, 69 % and 65 % respectively. In Chapter 4, it is proved that the accuracy of the introduced method is higher than the conventional algorithms and can be used dependably for first break calculation. It is also shown that the mis-pick elimination after SVMs classification results is an efficient and helpful tool which can be used in any dataset.