In the complex and rapidly changing environment of a battlefield, information is lost by factors such as interference of the enemy and limited sensor performance. To ensure that an unmanned aerial vehicle (UAV) can make threat assessments when sufficient information is lacking, this paper proposes a new Bayesian network (BN) learning method with a small dataset. For structure learning and parameter learning with a small dataset, the scoring function is constrained by the constraint matrix obtained by the Bootstrap method. The learning algorithm is proposed based on a BN structure learning with a small dataset and an interval a priori-constrained BN parameter learning algorithm. Simulation results demonstrated the higher accuracy and availability of the proposed method than traditional methods for UAV threat assessment on a small dataset.