We evaluated 3D inversion of land controlled-source electromagnetic (CSEM) data collected across the Ketzin CO2 storage formation. A newly developed, parallel and distributed 3D inversion code, which is based on a direct forward solver, has been used. This inversion scheme allowed us to calculate the Jacobian matrix explicitly within a reasonable time and use it to calculate regularization parameters, inspect survey coverage, and carry out resolution analysis. After demonstrating that the magnetic field components are sensitive to conductors only, whereas the electric field components are sensitive to all features of interest, we continued to work with electric field data only. Estimates of data uncertainty obtained from robust processing were used for automated data preselection and weighting during inversion. We tested different regularization techniques and a range of starting models to explore the model space and assess the influence of regularization on the inversion images. We further demonstrated an approach for handling numerical singularities due to sources located inside the inversion domain. We estimated survey coverage, horizontal and vertical resolution, and depth penetration using cumulative sensitivity, point spread functions, and depth-resolution plots. Based on data fit analysis, we determined a preferred subsurface conductivity model, which we compared to an independent regional structural geologic model, and we provided an interpretation for the structures resolved. The inversion approach we used provides robust results in good agreement with known geology, offers new possibilities for model assessment, and should be transferable to other CSEM data sets. [ABSTRACT FROM AUTHOR]