In a study of the detection of test item bias, Chinese speakers and Spanish speakers were administered a measure of pronunciation accuracy for the (d) and (z) morphemes. Six indices were calculated for each item: item difficulty, point biserial correlation coefficient, item variance, chi-square, Rasch logit of difficulty, and the average standardized difference between the expected probability of a correct response and the observed item response. Twelve different items were flagged as biased by one or more indices. Five items were flagged for both language samples. The classical item indices, item difficulty and the point biserial correlation coefficient, flagged more items than the remaining indices combined. This finding can be explained by the fact that both samples exhibited similar levels of performance on the pronunciation task. Analyses of variance indicated a significant effect for the choice of allomorph and the stem final place of articulation for the conditioning of the allomorphs. (Author/MSE)