BACKGROUND: Cognitive dysfunction is a major symptom in schizophrenia associated with social skills. It has been reported that cognitive rehabilitation can improve cognitive dysfunction. The Brief Assessment of Cognition in Schizophrenia-Japanese version (BACS-J) is often used as an outcome measure to assess the effectiveness of cognitive rehabilitation. However, the minimal clinically important difference (MCID) in the BACS-J composite score has not been reported. Therefore, we conducted this study to calculate a preliminary MCID in the BACS-J composite score and confirm the feasibility of retrospective data collection and analysis for future large-scale studies. METHODS: The medical records of patients with schizophrenia who underwent cognitive rehabilitation were retrospectively surveyed. BACS-J data were collected at the beginning and end of the cognitive rehabilitation, and Clinical Global Impression-Improvement (CGI-I) data obtained at the end of the cognitive rehabilitation were evaluated retrospectively. To calculate the MCID in the BACS-J composite score using distribution-based methods, the standard error of measurement was calculated as a characteristic of the scale itself. To calculate the MCID using anchor-based methods, the mean change in BACS-J score corresponding to "minimally improved" on the CGI-I was determined. RESULTS: Twenty-eight patients were included in this study, and BACS-J data were collected from all patients. The CGI-I was completed by 11 patients, 3 of whom showed "minimally improved" according to their CGI scores. Distribution-based methods applied to the data of 28 patients revealed an MCID of 0.735 for the BACS-J composite score. Anchor-based methods were ultimately not applied because the sample size was insufficient. CONCLUSION: This study confirmed that CGI and BACS-J data can be collected and analyzed retrospectively. According to distribution-based methods, an increase of approximately 0.7 in the BACS-J composite score can be considered clinically meaningful. Future studies with larger sample sizes using both calculation methods could provide more accurate MCID.