Abstract: The interval between a kick occurrence and a blowout is a critical window for preventing blowouts. To address the issues of delayed alarm times and high false alarm rates in existing Bayesian kick detection methods, improvements are necessary. This study analyzed the correlation between surface kick data to establish a tree-augmented Bayesian model. Sequential feature logging parameters were extracted from 91 drilling kick events in the Sichuan-Chongqing region to construct a training dataset. The trained model was tested to develop an early kick detection method based on the tree-augmented Bayesian network. Kick data from a well outside the training set was used as the test set for the detection model. The Bayesian-based kick detection model reduced the false alarm rate by 52.07% compared to other models. The detection model issued an alarm 16.6 minutes before the kick occurred, 510 seconds earlier than the naïve Bayesian early kick detection model. The tree-augmented Bayesian early kick detection model incorporates the correlation of anomalous parameters during kick events. It significantly advances the kick detection time while maintaining a lower false alarm rate. This approach provides a new framework for developing kick detection models based on large-scale logging data.