The extent, causes, and importance of context effects on item parameters for one- and three-parameter latent-trait models were examined. Items were taken from the California Achievement Tests Reading Comprehension and Mathematics Concepts and Applications subtests. The reading items were administered to 1,678 fourth-grade students, and the mathematics items were administered to 1,812 sixth-grade students. Context effects were measured by comparing the stabilities of item parameters obtained in the same context with the stabilities of parameters obtained in different contexts. Stability of item parameters was determined by testing the students twice with the same items. Changes in context were produced by changing the order of the items for the second test administration. Changes in context usually had substantial effects on the item parameters of both models, and to some extent these context effects appeared to be the result of the items' positions in the test booklets. Context effects appeared more important in item calibration than in item selection and in making predictions for single items than in making predictions or obtaining trait estimates with groups of items. (Author/MH)