Ultrasonic tomography (UT) is a nondestructive technique that is widely applied in industrial process imaging. In this paper, a novel sector diffusion method for construction of a sensitivity matrix is proposed to improve the performance of a linear back-projection (LBP) algorithm. The so-called sectordiffusion-matrix-based LBP (SLBP) was applied to reconstruct gas/liquid two-phase flow. A series of models has been established to evaluate the effectiveness of the method. The images reconstructed by traditional LBP and SLBP are presented and compared. Two quantitative evaluation criteria including position error and image correlation coefficient (I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">cr</sub> ) were defined and used to evaluate the quality of the reconstructed images. The imaging results indicate that the SLBP method is valid and feasible to improve the quality of reconstructed images, especially for a UT system with a small number of transducers. The evaluation criterion I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">cr</sub> is improved by 26%, 20%, 9%, and 4% for four different flow patterns with an eight-transducer UT system in 128 × 128 pixels and improved by 26%, 20%, 10%, and 1% in 40 × 40 resolution.