First-arrival picking is an essential step in seismic data processing， which can directly affect the accuracy of normal moveout correction， static correction and velocity analysis. At present， affected by background noise and complex near-surface conditions， the picking accuracies of the existing methods are reduced. Based on this， a First-arrival automatic Picking algorithm based on Clustering and Local linear regression （FPCL） was proposed. This algorithm was implemented in two stages： pre-picking and fine-tuning. In the pre-picking stage， the k-means technique was firstly used to find first-arrival cluster. Then the Density-Based Spatial Clustering of Applications with Noise （DBSCAN） technique was used to pick first-arrivals from the cluster. In the fine-tuning stage， the local linear regression technique was used to fill in missing values， and the energy ratio minimization technique was used to adjust error values. On two seismic datasets， compared with Improved Modified Energy Ratio （IMER） method， FPCL had the accuracy increased by 4.00 percentage points and 3.50 percentage points respectively； compared with Cross Correlation Technique （CCT）， FPCL had the accuracy increased by 38.00 percentage points and 10.25 percentage points respectively； compared with Automatic time Picking for microseismic data based on a Fuzzy C-means clustering algorithm （APF）， FPCL had the accuracy increased by 34.50 percentage points and 3.50 percentage points respectively； compared with First-arrival automatic Picking algorithm based on Two-stage Optimization （FPTO）， FPCL had the accuracy increased by 5.50 percentage points and 16.25 percentage points respectively. The above experimental results show that FPCL is more accurate.