For complex educational assessments, there is an increasing use of item families, which are groups of related items. Calibration or scoring in an assessment involving item families requires models that can take into account the dependence structure inherent among the items that belong to the same item family. This article extends earlier works in three directions: (a) extends the model to take into account item families with polytomous items and implements a Markov chain Monte Carlo algorithm for the estimation of the model parameters, (b) generalizes family response functions to polytomous item families and defines the family score function as ways to examine each family graphically, and (c) uses Bayes factors to select either the more complicated model or a simple model. All three extensions of the earlier works on item families are demonstrated using two data sets: one simulated and one from the National Assessment of Educational Progress.