Human middle ears show large morphological variations. This could affect our perception of hearing and explain large variation in experimentally obtained transfer functions. Most morphological studies focus on capturing variation by using landmarks on cadaveric temporal bones.We present statistical shape analysis based on clinical cone beam CT (CBCT) scans of 100 patients. This allowed us to include surface information on the incudomallear (IM) complex (joint, ligaments and tendon not included) of 123 healthy ears with a scanning resolution of 150 μm and without a priori assumptions. Statistical shape modeling yields an average geometry for the IM complex and the variations present in the population with a high precision.Mean values, variation and correlations among anatomical features (length of manubrium, combined length of malleus head and neck, lengths of incus long and short process, enclosing angles, ossicular lever ratio, incudomallear angle, and principal moments of inertia) are reported and compared to results from the literature. Most variation is found in overall size and the angle between incus and malleus. The compact representation provided by statistical shape modeling is demonstrated and its benefits for surface modeling are discussed.