The artificial rabbits optimization(ARO) has some disadvantages such as low precision, slow convergence, and easy to fall into local optima. A specular reflection learning based ARO (SRL-ARO) algorithm was proposed. In SRL-ARO, an adaptive energy decreasing factor was designed to ensure a smooth transition of the algorithm from exploration to exploitation. The global optimum information was embedded in the exploitation phase to speed up convergence and improve solution precision. The specular reflection learning strategy was introduced to enhance the population diversity and jump out of local optima. Six benchmark test functions were selected for experiments. The results show that the proposed SRL-ARO algorithm shows significant improvement in terms of accuracy and convergence compared with other intelligent optimization algorithms. Finally, SRL-ARO was employed to deal with the feature selection problem. The compared findings indicate that the total performance of the proposed SRL-ARO is superior to other methods.