In order to improve the defect of insufficient data utilization in BN structure learning under small data sets，meanwhile，to improve the optimization efficiency of the Bayesian structure learning algorithm，the improved Bayesian network based on improved ant lion optimization structure learning algorithm (ISB-ALO) is proposed.Firstly，the network is Initially constructed through mutual information constraint，and the ant lion algorithm is initialized.Secondly，in order to effectively use the small data set，the matrix elements in the iteration are converted with improved sigmoid function.To further improve the search efficiency of ant lion algorithm，replace individual ants with the migration，variation and clearing operator in the biogeographic algorithm；Finally，combining update mechanism according to ant lion algorithm looking for the optimal solution.The experimental results show that the ISB-ALO has high optimization efficiency and fast convergence speed,which can jump out of the local optima and have higher accuracy.