Model-based control can provide high-accuracy performance over position-based or velocity-based control. Therefore, to employ model-based control in industrial robots, it is important to estimate the dynamic parameters as accurately as possible. However, traditional estimation methods, such as least squares (LS), are not sufficiently accurate, and the feasibility of dynamic parameters cannot be guaranteed. In this article, an iterative hybrid least square (IHLS) algorithm is proposed to estimate the base parameters of industrial robots by dividing the identification processes into two loops. The inner loop integrates a linear matrix inequality with the semidefinite programming technique to guarantee physical feasibility and reuses the torque deviations between the measured torque and predicted torque to estimate the base parameters of the robot, while the outer loop substitutes the Stribeck friction model for the Coulomb-viscous friction model to estimate the joint friction torque. Moreover, backpropagation neural network (BPNN) is introduced to further estimate the joint friction torque based on the Stribeck friction model. Experiments are conducted on two industrial robots, and four methods are compared in dynamic parameter identification. Experimental results show that the hybrid approach of the IHLS algorithm with the BPNN has the best performance among the four methods.