ABSTRAKObjectives of this study are: 1) to identify gender bias in science items, (2) to determine the most sensitive method for detecting item bias by considering both the internal and external factors (3) to explain contribution of external factors to item bias through a qualitative analysis of biased items.This study represents substantial contribution to knowledge in that it indicates the importance of detecting bias in items .Since item bias could produce either an overestimate or underestimate measurement resuits; it is obviously a serious matter of fairness in testing, especially when a certain group or some people are disadvantaged.In this study, the science test items from the "Trends In International Mathematics and Science Study? (TIMSS) 2003 were analysed using a structural equation model approach under the LISREL program.It is found that some science items used in the TIMSS 2003 are biased. Furthermore, the uses of structural equation model could make it possible not only to identify items which are biased against boys or girls, but also to provide explanatory information in regard with its characteristics.