The purpose of this study was to investigate the effects that test directions of multiple choice tests produce on examinees' test-taking behaviors and performance. The data for this research were obtained by having students respond to 57 items of Shiba's Word Meaning Comprehension Test that constitutes the version U2. Three forms of U2 were administered under number-right scoring directions, formula-scoring (correction-for-guessing) directions, and neutral directions. There were many omitted responses by examinees under formula scoring directions. On the other hand, there were few omissions under number-right-directions. Three treatments of omitted responses were examined in item response theory: treating as wrong, ignoring, and Lord's method. Great differences existed in estimated ability parameters among these treatments under formula-scoring directions, differences under neutral directions were minor, and one under number-right directions was minimum. It was concluded that formulascoring directions should not be used, since formula-scoring directions restrict examinees' test-taking behaviors and their performance is greatly affected by treatment of omitted responses under such directions.