The analysis of fit of data to a measurement model for graded responses is described. The model is an extension of Rasch's dichotomous model to formats which provide more than two levels of response to items. The model contains one parameter for each person and one parameter for each "step" in an item. A dichotomously-scored item provides only one step from fail to pass so only one item parameter is estimated. An item with three ordered performance levels (0, 1, and 2), provides two steps. For this item two parameters are estimated: the first governs the model probability of scoring 1 rather than 0, and the second governs the probability of scoring 2 rather than 1. The model probability of completing the k'th step in an item is described by a simple logistic ogive of slope one and is governed by the person's ability and the difficulties of the step. This "Partial Credit" model's procedures for estimating the parameters are unconditional maximum likelihood. Both the effects of individual differences in response style and variations in the fit of items are examined, including items with large positive or large negative standardized mean squares.