Picking the first arrivals is an important step in the near-surface modeling method. The precise picking of the first arrivals can affect many aspects of the work, such as static correction and velocity modeling, etc. However, the first arrivals are affected by strong background noise and complex near-surface structure, which lead to the reducing accuracy. This paper proposes the first-arrival automatic picking algorithm based on two-stage optimization(FPTO). First-arrival automatic picking algorithm based on two-stage optimization divides the picking question into two sub-questions. According to these two sub-questions, two optimization targets are generated respectively. Next, two optimization functions are generated by these two optimization targets. First-arrival automatic picking algorithm based on two-stage optimization takes two steps: the first step is to use the vertical sliding window to find out the possible range of the first arrivals; the second step improves the energy ratio algorithm and picks up the first arrivals in the range found in the first step. Compared with three algorithms(Backpropagation neural networks method，BNN；Direct correlation method，DC；Modified Coppens’s method，MCM) on the two data sets, experimental results show that the accuracy and the stability of the FPTO algorithm are greatly improved.