Pennation angle is the most often adopted sonomyography (SMG) measurement to quantify muscle activities under different contractions because of its independence on the position and the field of view in B-mode imaging. In this paper, a new paradigm for automatic pennation angle estimation was described, which used some unique features of musculoskeletal ultrasound images, including parallel fascicle pattern, and hyper-echogenic band for fascicles and aponeuroses. The automatic algorithm is comprised of three stages. Sticks enhancement followed by an anisotropic diffusion filter was used to enhance strip-band pattern and restrict speckles in the ultrasound image. And then the normalized Radon transforms was implemented to detect the aponeuroses and identify the fascicle region in the ultrasound image. Finally, average value of local maximum points in each orientation was calculated in Radon spaces. The dominant fascicle orientation was confirmed by searching the maximum average value, called consistence voting. This detection made use of the texture information of parallel fascicle pattern to detect its orientation instead of single line feature extraction like the localized Radon transform and revoting Hough transform reported previously. The pennation angle was then obtained as the difference between the fascicle orientation and aponeuroses orientation. The algorithm was evaluated using synthetic images with various noise levels and real musculoskeletal ultrasound images of gastrocnemius muscles. The experiment results showed that both our proposed method were robust to noise in images. The mean absolute difference between the proposed method and manual measurement for pennation angle was 0.71 degree. It was demonstrated that the proposed method for automatic measurement of pennation angle in ultrasound images of muscle was reliable. The new algorithm may be beneficial to the quantitative evaluation of muscle functionality with sonomyography in the human motion anal...