The purpose of this research was to reinvestigate the accuracy of three item bias detection procedures: (1) Linn and Harnisch's pseudo-IRT(Z) method; (2) Camilli's chi-square technique; and (3) Angoff's revised transformed item difficulty method. These methods are applied when the minority group sample size is too small to obtain stable estimates of item parameters. This study analyzed the data which included ten black slang items imbedded within a standardized vocabulary test. In order to determine the best methodology, three statistics were calculated: a point biserial correlation between an a priori bias index and the detected bias index associated with each method, intercorrelations among the bias measures of three procedures, and the percentage of agreement between the a priori bias index and bias index based on each method. Results showed that (1) the chi-square technique is slightly more accurate than the pseudo-IRT(Z) method in detecting bias; (2) Angoff's revised transformed item difficulty (TID) method is considerably worse; and (3) the chi-square procedure is highly correlated with the pseudo-IRT(Z) method. Appendices include item bias indices for all items and all methods, item information for computing the item bias index of Angoff's revised TID method for white and black groups, estimates of item parameters based on the three-parameter logistic model for Linn-Harnisch's method, and principal component analysis of the test item. (Author/JAZ)