Magnetotellurics (MT) is now an important geophysical tool for imaging complex geothermal energy systems, but retrieving realistic three-dimensional (3D) resistivity models is challenging, and computationally intensive. To ameliorate this situation, we develop a sequenced MT inversion workflow that formalizes and improves upon concepts introduced by Rosenkjaer and Oldenberg (2012). We test the performance and reliability of our workflow using five 3D MT datasets from geothermal fields in the United States (Coso, Raft River) and Iceland (Krafla, Hengill and Krysuvik). We compare inversion results from the standard and sequenced workflows for each experiment, and find that the sequenced workflow always retrieves a better-fitting model, specifically at lower frequencies, which is geologically consistent and representative of a high temperature geothermal system. It also uses fewer computational resources. Published by Elsevier Ltd.