Whether item response theory (IRT) is useful to the small-scale testing practitioner is examined. The stability of IRT item parameters is evaluated with respect to the classical item parameters (i.e., p-values, biserials) obtained from the same data set. Previous research investigating the effect of sample size on IRT parameter estimation has usually been performed on simulated data. The present study follows a few others in using a real-life small-sample testing application. The procedure involved obtaining a common set of items administered to three small-sample groups of examinees over a 3-year period (sample sizes of 173 in 1988, 149 in 1989, and 106 in 1990, respectively) and estimating the item parameters with both restricted and unrestricted IRT models. The test was a national certification examination for certified public accountants wanting certification in personal financial counseling. Classical reliability (p-value) and biserial correlation analyses were performed prior to traditional one-, two-, and three-parameter IRT analyses. Results suggest that stable item difficulty parameters can be obtained for small sample sizes using the one-parameter or modified two-parameter model when the data fit the IRT model (i.e., when they are unidimensional). The IRT and classical analyses performed could not successfully provide stable item discrimination parameters. However, the conditions under which IRT is useful to the small-sample test practitioner are discussed. Eleven tables present study data. A 24-item list of references is included. (SLD)