Early identification of students with learning difficulties is important for early intervention to prevent future failure. This article reported the investigation of the rating scale property, difficulty level, and dimensionality of a 30-item observation checklist for teachers (OCT-30), its subsequent revision, and the validity of the revised version. The sample included 1966 primary one students randomly selected from 51 randomly selected schools. Teachers of these 1966 students completed the OCT-30 at the end of the first semester. Based on their OCT-30 scores, 91 students out of the 1966 participants, with 31 from the bottom 5%, 30 from the bottom 6% to 20%, and 30 from the remaining 80%, were selected through random sampling for further assessment by educational psychologists. Rasch analysis indicated that the infit and outfit statistics of seven items of OCT-30 were outside the recommended range. These items were removed, resulting in a 23-item scale (OCT-23). Investigation of category functioning indicated that teachers could meaningfully differentiate between the categories. Item map analysis indicated that the OCT-23 was relatively easy for the participants, which was in line with the purpose of the checklist in screening students with learning difficulties. Among the 91 students assessed by educational psychologists, the total scores of OCT-23 were consistent with the judgment of educational psychologists. Confirmatory factor analysis supported the originally postulated 4-factor model. The reliability was 0.95. The revised checklist (OCT-23) demonstrated satisfactory psychometric properties in terms of reliability and validity.