The invention discloses a gravity data and magnetotelluric data joint inversion method. The method comprises a data collection step, a data screening step, an initial model determination step and a joint inversion step. By taking data quality as a standard, an initial model during joint inversion iteration is selected. Compared with a uniform semi-space model given in a unified way in the prior art, the method has the following beneficial effects that the joint inversion efficiency is improved, the inversion calculation time is shortened, the problem of iterative misconvergence caused by dataerrors is avoided, and the precision of a joint inversion result is improved.