We aimed to identify which of the 17 items comprising the Children's Depression Rating Scale-Revised (CDRS-R) can sensitively capture changes in depression severity. We used data from four studies involving two antidepressants. For each of the 17 CDRS-R items, we conducted item response analyses to identify and evaluate those that reflect changes in depression severity. We created plots of the item characteristic curves (ICCs) estimated by the graded response model, and option characteristic curves and ICCs using nonparametric item response theory. The change from baseline in the CDRS-R subscale score with specified reflective items by item response analyses and the effect size between the treatment group and placebo group were calculated and compared with those of the CDRS-R total score. CDRS-R items #2 (difficulty having fun), #3 (social withdrawal), #10 (low self-esteem), #11 (depressed feelings), and #15 (depressed facial expression) have favorable profiles that reflect disease severity. Changes from baseline in the CDRS-R total score (least square mean ± standard error) at week 8 were -22.3 ± 0.7 and -23.9 ± 0.7 in the placebo group and treatment group, respectively (difference, -1.5; estimated effect size, -0.113), changes from baseline in the CDRS-R5 (CDRS-R subscale consisting of the specified reflective items [#2, #3, #10, #11, and #15]) score at week 8 were -8.4 ± 0.3 and -9.6 ± 0.3 in each group, respectively (difference, -1.2; effect size, -0.202). The item response analyses clarified the properties of 17 items of the CDRS-R for major depressive disorder in children and adolescents. The CDRS-R5 might optimize the assessment of changes in overall depression severity and differentiation of treatment responses.