The improved backtracking search algorithm based on oppositionbased learning and orthogonal crossover is proposed to overcome the problem of the slow convergence speed and sinking into local optima in the inversion of atmospheric duct. In the proposed algorithm, the oppositionbased learning is adopted to produce the better initial population, while the orthogonal crossover helps to avoid local optima and enhence the global search ability as well as the accuracy. The benchmark function and inversion problem of atmospheric duct are used to test the performance of the algorithm. The results show that the proposed algorithm has better convergence speed and accuracy.