A computationally efficient procedure for identifying the structure and parameters of a linear discrete-time SISO system is presented. The procedure is based upon a new computational technique of parameter estimation, using the instrumental variables, that eliminates a number of redundant computations and does not require explicit evaluation of the instrumental variables. The structure determination is aimed at providing a combination of low residual error, low output error and low parameter estimation error. Application of the proposed procedure on simulated, as well as real, data is reported.