An optimization technique consisting of a Levenberg-Marquardt algorithm is coupled with a distributed rainfall-runoff model. The purpose is to estimate the rainfall-runoff model parameters appropriate for the Yuragawa-river flood by Typhoon No.23 on 19-22 October 2004. The application exhibits that the simulation-optimization model can not only estimate the model parameters well but also quantify sta tistically the relations among the model, the observations and the parameters. This paper describes the usefulness of the relative composite parameter sensitivty (RCPS), the composite observation sensitivity (COS), and the parameter coefficient matrix.