It was found that the results of the RELAP-3 code, which is one of the typical analytical codes for analysis of the loss-of-coolant accident (LOCA) of light water reactors, do not agree well with the results from the ROSA-I experiments under certain break conditions. It was determined that the discharge coefficient used in the code as a parameter can be correlated with the quality of the discharged fluid and that the calculated liquid mass transient does not always agree with the experimental one when a constant bubble escape velocity is assumed. These difficulties come from the possibility of shortcomings of the model dealing with the LOCA phenomena. An attempt was made to improve these aspects of the code by incorporating the correlation of the discharge coefficient with the quality and with Wilson's experimental interpretation of the bubble velocity and certain assumptions in its application. The results obtained by the modified code are in good agreement with both those from the ROSA-I experiment and experiments at Hitachi Ltd.