X-ray spectrum plays an important role in dual spectral X-ray computed tomography (CT) imaging, CT beam hardening correction, quantitative CT analysis and so on. The conventional methods estimate the spectrum from a set of transmission data measured directly for different thicknesses of step-wedge phantoms. A novel estimation method of X-ray spectrum is proposed. The proposed method has two features. First, the dependence of the transmission intensity on the thickness of attenuation materials from the CT data of simple phantoms is restored. This not only simplifies the phantom production and the measurement process, but also can restore a more accurate dependence of the transmission intensity on the thickness of materials. Second, an alternative iteration algorithm for the spectrum estimation based on the parameter spectrum model is developed, in which the effect of the anode material is considered. The algorithm can relieve ill-posedness and instability for solving the spectrum estimation problem, by using the parameter X-ray spectrum model as constraint condition. The results of numerical experiments with several simulation data suggest that the proposed method is capable to reconstruct X-ray spectra more accurately and robustly for both bremsstrahlung and characteristic photons, compared with some transmission measurement methods.