Abstract: Clinical data are abundant, but there are some problems that limit clinicians' use of these data for clinical research, such as incomplete baseline data, confounding bias due to nonrandom grouping, and higher loss of follow-up. The inherent defects of retrospective design can be overcome by limitations, matching, stratification, multivariate analysis, propensity score, and other methods. This article elaborated how to do a high-quality cohort study from the perspectives of design, analysis and paper writing.