The invention provides a simple and convenient nuclear magnetic resonance gradient waveform distortion pre-correction method. The method comprises the following steps: 1, correcting: loading an ideal gradient waveform, namely Sgg (i = 1, t) = Sideal (t), in a gradient generator, generating distortion after the gradient waveform passes through a gradient system, obtaining a gradient waveform Svol (i = 1, t) in an imaging space through a sequence method, and estimating a gradient system response according to the Svol (i = 1, t) and the Sideal (t) to obtain a system impulse response h (t); correcting distortion of the Svol (i = 1, t) relative to the Sideal (t), performing next iteration, and updating an output gradient waveform Sgg (i + 1, t); and iterating for multiple times until the distortion of the Svol (i = 1, t) is minimum, and repeating the following steps to obtain a pre-corrected gradient waveform database. The method does not need to specially measure the frequency response of the gradient system, is more stable and reliable, does not need pre-scanning, and can adapt to scanning in any geometric orientation.