Taiwan has implemented the National Health Insurance (NHI) program, and DRG-based prospective payment system will be adopted. Under DRG/PPS, the coding of disease classification certainly relates directly to the payment rate. To transit from fee-for-service system to prospective payment system, it is imperative to evaluate the coding quality of disease classification, to understand its influencing factors, and to study how to improve the quality of coding. The study used a stratified sample of 1,300 inpatient cases from the NHI claim data in September, 1995. The discharge summary of these cases were reviewed and recoded by coding specialists to compare with the original coding. The characteristics of the original coders were also collected in order to explore the related factors affecting the quality of disease classification. The study found that there were 63%of the study cases with at least one coding error. The average rate and number of miscoding for each case was 35.56% and 1.42, respectively. The overall rate of coding errors of the whole sample was 52%. Among the errors, 33% occurred in the principal diagnosis, and 39% were due to coding errors in the principal procedure. Ownership, accreditation level and the scale(number of beds) of hospitals were found to be significantly important in determining coding quality. The coding quality of the proprietary hospital was better than that of those public and private hospitals. However, there was no significant difference in coding quality between public and private hospitals. Both medical centers and regional hospitals had better coding quality than district hospitals did, whereas no significant difference was found between med ical centers and regional hospitals. In addition, coding qualification of coders, the work load of coders per month, and the time devoted to disease coding per day were also important determinants of coding quality.