An image reconstruction approach for near-infrared optical tomography (NIR OT) based on a parallel neural network is presented. The parallel BP neural network is used to distinguish the nonlinear relationship between the spatial location of tumor and light intensity around the boundary of tissue. The method turns a complicated model into two simpler ones to build two parallel BP neural networks. The steady state diffusion equation of the two simple models is solved by Femlab software. The inverse problem is solved as optimization problem by Levenberg-Marquardt algorithm. The concept of the average optical coefficient is proposed, which is helpful to understand the distribution of the scattering photon from tumor. The reconstruction can be obtained by the trained network. The fast reconstruction of tissue optical properties and provided reconstruction of OT with a new method can be realized by proposed algorithm.