Abstract: A systematic global identifiability analysis method of large and complex models was proposed. The method was based on the global sensitivity, Fisher matrix and the normalized Fisher information matrix and could take the effect of constraint into account. The identifiability measure includes sensitivity index, collinearity index and determinant index. Firstly the statistical approach was applied to integrate the local matrix into the corresponding global matrix. The constraint was introduced into the objective function through the penalty function method and the penalty coefficient was determined by the global competitive ranking method. Secondly, the global identifiability analysis on the macro-micro coupled viscoplastic constitutive model considering dynamic recrystallization was taken. The result shows that the identifiability is poor due to high correlation coefficient between some parameters.