Aiming at improving the conflict detection precision in a single classifier, a probabilistic conflict detection algorithm based on ensemble learning was proposed. Firstly, aircraft conflict model was established for the purpose of selecting flight datasets. Secondly, the current positions, speed vectors, look-ahead time, estimated turning time and turning angles were extracted as characteristic quantities which were inputted to train the basic classifiers, and a meta-datasets were obtained. Afterwards, support vector machine is used as the second-level classifier, the meta-datasets was regarded as new characteristic quantities which is used to train the stacking meta classifier. Lastly, the conflict probability was solved by the Sigmoid function mapping method. Simulation results have shown that this algorithm has a high accuracy for conflict detection and is suitable for turning flight.