Appropriate estimates of earthquake response spectrum are essential for design of new structures, or seismic safety evaluation of existing structures. This paper presents an alternative procedure to construct design spectrum from a combined normalized response spectrum(NRSC) which is obtained from pseudo-velocity spectrum with the ordinate scaled by different peak ground amplitudes(PGA, PGV, PGD) in different period regions. And a consecutive function f(T) used to normalize the ordinates is defined. Based on a comprehensive study of 220 strong ground motions recorded during recent eleven large worldwide earthquakes, the features of the NRSC are discussed and compared with the traditional normalized acceleration, velocity and displacement response spectra(NRSA, NRSV, NRSD). And the relationships between ground amplitudes are evaluated by using a weighted mean method instead of the arithmetic mean. Then the NRSC is used to define the design spectrum with given peak ground amplitudes. At last, the smooth spectrum is compared with those derived by the former approaches, and the accuracy of the proposed spectrum is tested through an analysis of the dispersion of ground motion response spectra. Appropriate estimates of earthquake response spectrum are essential for design of new structures, or seismic safety evaluation of existing structures. This paper presents an alternative procedure to construct design spectrum from a combined normalized response spectrum (NRSC) which is obtained from pseudo-velocity spectrum with the ordinate scaled by different peak ground amplitudes (PGA, PGV, PGD) in different period regions. And a consecutive function f (T) used to normalize the ordinates is defined. Based on a comprehensive study of 220 strong ground motions recorded during recent eleven large worldwide earthquakes, the features of the NRSC are discussed and compared with the traditional normalized acceleration, velocity and displacement response spectra (NRSA, NRSV, NRSD). And the relationships between ground amplitudes are evaluated by using a weighted mean method instead of the arithmetic mean. Then the NRSC is used to define the design spectrum with given peak ground amplitudes. At last, the smooth spectrum is compared with those derived by the former approaches, and the accuracy of the proposed spectrum is tested through an analysis of the dispersion of ground motion response spectra.