This paper presents an investigation of a method for assessing the goodness-of-fit of IRT models using estimated item-observed score regressions. An advantage of this method is that examinees are assigned to cells on the basis of an observed quantity, their number-right score. The method for assessing goodness-of-fit currently used assigns examinees to cells on the basis of an estimated quantity, their estimated ability. Comparisons are made between the two methods using simulated and actual data. A disadvantage of the new method is that it cannot easily handle tests where many examinees do not finish. (68pp.)