High-quality interpretation of seismic refraction data depends on an accurate and reliable identification of the first arrival times. The first arrivals can be achieved through a graphic or image by conventional picking, but this process depends on external factors such as scale and quality of imaging of data, amplitude ratio, sensitivity of picking cursor and user experience. Under these considerations, defining of first arrivals on noisy data becomes more complex and instable. In this study, Cross-Correlation Technique (CCT), which is widely used in the process of reflection data, has been used in picking of the first arrival times on noisy or noiseless seismic refraction data by a semi-automatic process. The CCT has reduced dependence on user and decreased incorrect picking caused by environmental noise, display characteristics and scaling factor. The CCT has been tested on synthetic models with different noises content and various field data. Chi-square error criterion was assessed to performance of the pickings. In addition, effects of small-time differences between the conventional picking process and CCT have been demonstrated on a refraction tomography velocity section. Therefore, we believe that our proposed method is a useful contribution to the existing methods of first arrival picking.