A model of a walking pattern imitating human motion is presented. An accurate imitation of human motion and a robust bipedal walking motion are, however, hardly realized together. We therefore focus on only three characteristics of human walking motion: single toe support, knee stretching, and swing leg motion. Based on a conventional pattern generator, single toe support is added, waist height is changed in order to stretch the knees as much as possible, and swing leg motion is generated approximating the human's motion. The generated motion is then filtered to provide a feasible pattern. In addition, the stabilizer is improved in order to keep the Zero Moment Point (ZMP) within the tiny support polygon during the single support phase with toe link. Finally, we successfully demonstrate the generated walking pattern with the robot HRP-4C.