Abstract: The algorithm and the computer program TFIP used to identify the transfer function models from experimental freguency response data are presented. It is also considered how the selection of the model orders, the selections of the equationerror and real error tolerance, and the modes for updating weight coefficientsaffect the convergent speed and identification results. The new term rejectioncriterion and the method for updating weight coefficients by variable mode are proposed, the former can simplify the program, and the latter may greatly accelerate the convergence of the weight coefficients.