The purpose of this study was to compare empirical and subjective procedures for identifying science test items which discriminate against males or females in a fifth grade population (n=926). Techniques such as the Mantel-Haenszel procedure enable test developers to use both a priori and ex post facto procedures to identify specific items that may be functioning with gender bias. The test used in this study was developed as an end-of-year, criterion-referenced instrument and was based on the standards and objectives in the Utah State Science Core Curriculum for the fifth grade. All items were multiple-choice; some used illustrations, but most were text only. The test was initially administered to approximately 300 fifth-grade students. Items were compared that were identified as functioning differentially toward gender. Results reported in this study include: (1) the two methods used led to different results; and (2) several items identified by these methods were more difficult for females. Suggestions for the use of each technique are provided, together with the characteristics of items that discriminated on the basis of gender. (CW)