An artificial neural network method is proposed for estimating reduced scattering coefficient μ_s' and phase function parameter γ of biological tissues from spatially resolved reflectance profiles in the sub-diffusive regime. Monte Carlo simulation method is used to obtain data samples of diffuse reflection from biological tissues. These data samples are used to train back-propagation neural network get the information of γ predicted from the sub-diffused scattered light. Since there is a large error occurs when predicting μ_s' and γ simultaneously, the segmenting data train of two back-propagation networks is performed to identify the μ_s' and γ in turn. It is found that 3.64l_th (l_th representing the average transport free path) is an insensitive points of γ. The network trained with data samples near this point is used for predicting μ_s', while the network trained with data samples in the 2l_th is used for predicting γ. Monte Carlo simulation result show that within the range 1.3 ≤ γ ≤ 1.9, the relative root mean square error between the predicted result and the true value is within 1%. Compared with the existing measurement methods, the proposed method is simpler and has improved accuracy.