Detecting and localizing tumors by sensing the mechanical property of soft tissue is useful in disease screening and surgical planning. Examining and extracting human subjects' policies in soft tissues palpation tasks can aid in developing autonomous palpation systems. This paper aims at introducing a programing by demonstration approach to encode the underlying features in human demonstration and generalize a smoothed trajectory for robot reproduction. Specifically, demonstrations of the designed palpation task from different human subjects are recorded. GMM is used for motion modeling and GMR is used for trajectory generation. A UR5 robot automatically reproduces the palpation task with the generated trajectory. Though only trajectory data are utilized for learning and reproduction, force profiles are recorded for segmentation and task evaluation. A video showing human palpation as well as robot reproduction is available at: https://youtu.be/c4xEjyQZIvU.