The problem of automating preoperative planning of the femoral component (stem) for total hip arthroplasty (THA) is addressed. In our previous method, time-consuming trial-and-error processes were involved in parameter tuning of the objective function. This problem prevents application in different stem systems. To overcome this problem, a statistical surgical plan atlas (SSPA) is constructed from training datasets of stem planning. The SSPA represents the average and variance of the distance distribution on the stem surface to the femoral canal surface. That is, it encodes the distribution of the degree of contact preferred by the surgeon. Automated planning is performed by minimizing the squared difference between distributions of the SSPA and planning solution. The proposed method involves no parameter tuning to define the objective function that evaluates differences from the planning the surgeon prefers. Experimental evaluations showed that the proposed method renders parameter tuning unnecessary while it still provides comparable accuracy to the previous method.