Abstract: A postnonlinear unmixing algorithm was presented for hyperspectral images based on backtracking optimization to improve the unmixing accuracy. On the basis of the postnonlinear mixing model, the reconstruction error between the observed images and the reconstructed images was used as the objective function, backtracking search optimization algorithm was used to search in the solution space to obtain the optimal solution which minimize the objective function. In the search process, the boundary control mechanism of the backtracking search optimization algorithm effectively ensured the constraint condition in the hyperspectral image unmixing, and then the abundance and nonlinear parameters can be estimated accurately. The experiments conducted for both synthetic images and real remote sensing images show that the algorithm proposed is provided with excellent unmixing performance. The unmixing accuracy achieved is significantly better than the state-of-the-art nonlinear hyperspectral images unmixing algorithms.