A Finite element (FE) model of a human middle ear is developed, assessed, and updated using a statistical approach. The model consists of three ossicles (malleus, incus, and stapes), a tympanic membrane, tendons, and ligaments. The uncertainty of the model input parameters associated with the material properties and boundary conditions are considered in order to assess the validity of the model. The variation of the umbo displacement transfer function (UDTF) as a result of the uncertainty of the model input parameters is estimated and compared with those from experiments. Using the analysis of variance (ANOVA) with a three-level orthogonal array, the most important calibration parameters, which are composed of stiffness-related and density variables, are selected. Furthermore, a metric for statistical calibration is introduced. Through minimizing the calibration metric, the calibration parameters are updated in order to enhance the performance of the middle ear FE model. The proposed statistical calibration framework effectively improves the middle ear FE model performance.