Impedance inversion is one of the most efficient tools to predict and characterize oil and gas reservoirs. Small faults have the ability to reveal the high-resolution information on the target reservoir. However, the impedance information of small faults is always hidden during standard inversion process. Instead, diffracted wavefield indicate high-resolution information beyond reflected wavefield, yet it is biased by reflection-oriented migration. Usually, in the standard inversion scheme, the input is the migrated data and the impedance of small faults is distorted. Therefore, a novel approach is proposed to improve impedance inversion results using the unmigrated data and diffracted wavefield. The input data for inversion is a zero-offset section with identified diffraction events. Forward modelling, designed for the impedance inversion, includes the classical specular reflection plus asymptotic diffraction modelling schemes. A synthetic model, with the fault of half of the wavelength, shows the validation of this novel approach to improve the resolution and accuracy of impedance inversion.