This paper redefines the term "test bias," and the basic assessment model. It enumerates the more common causes of test bias and discusses ongoing research (especially at ETS) designed to alleviate its effects. Distinction is made between two types of test bias: psychometric bias, which is primarily due to qualities inherent in the test itself, and prediction or selection bias, which is more likely to be due to how the test is being used. Tests which systematically measure certain subgroups of the population with greater error are said to be "psychometrically biased," or to have "differential reliability." Tests which systematically predict achievement criteria with greater errors for certain subgroups than for others are said to have "predictive bias" or have "differential validity." It is suggested that psychological factors, in particular motivation, may be the more significant contributor to findings of predictive bias, thus rather than look for racial difference in "test bias" we might pay more attention to more general motivational factors which may cut across racial groups or sub-cultures within our applicant population.