Traction force microscopy is a useful technique for measuring mechanical forces generated by cells. In this method, fluorescent nano beads are embedded in the elastic substrate of cell culture, on which cells are cultured. Then, cellular forces are estimated from bead displacements, which represent the force-induced deformation of the substrate under the cell. Estimating the forces from the bead displacements is not easy when the bead density is low or the locations of cellular attachments are unknown. In this study, we propose a Bayesian algorithm by introducing a prior force direction that is based on cellular morphology. We apply the Bayesian framework to synthetic datasets in conditions under which the bead density is low and cellular attachment points are unknown. We demonstrate that the Bayesian algorithm improves accuracy in force estimation compared with the previous algorithms.