Abstract: Missing header earthquake records account for approximately half of the strong motion observation data.In P wave parameter related algorithm,it is necessary to automatically remove these recordings from large volume of observation data.This study proposes an automatic identification model of missing header earthquake records based on the eXtreme Gradient Boosting Tree (XGBoost) algorithm.The XGBoost model was established using 83 825 vertical component acceleration records from 970 earthquakes recorded by the K-NET in Japan.The results show that recognition success rate was 92.07% for positive samples (complete earthquake records) and 98.93% for negative samples (missing header earthquake records).Compared with Fisher Linear Discriminant model based on the same test dataset,our proposed model greatly improves the recognition success rate of positive samples,and keep high recognition success rate for negative samples.The results reveal that our model have high recognition accuracy for recordings with (without) missing header.Furthermore,when automatically extracting P-wave parameters from vast strong motion observation data,the model may be utilized to automatically pick and exclude recordings with missing header earthquake,avoiding these recordings degrading the data quality.