An overview of the whole process of reconstructing the coronal magnetic field from boundary data measured at the photosphere is presented. We discuss the errors and uncertainties in the data and in the data reduction process. The problems include noise in the magnetograph measurements, uncertainties in the interpretation of polarization signals, the 180° ambiguity in the transverse field, and the fact that the photosphere is not force-free. Methods for computing the three-dimensional structure of coronal active region magnetic fields, under the force-free assumption, from these boundary data, are then discussed. The methods fall into three classes: the ‘extrapolation’ technique, which seeks to integrate upwards from the photosphere using only local values at the boundary; the ‘current-field iteration’ technique, which propagates currents measured at the boundary along field lines, then iteratively recomputes the magnetic field due to this current distribution; and the ‘evolutionary’ technique, which simulates the evolution of the coronal field, under quasi-physical resistive magnetohydrodynamic equations, as currents injected at the boundary are driven towards the observed values. The extrapolation method is mathematically ill-posed, and must be heavily smoothed to avoid exponential divergence. It may be useful for tracing low-lying field lines, but appears incapable of reconstructing the magnetic field higher in the corona. The original formulation of the current-field iteration method had problems achieving convergence, but a recent reformulation appears promising. Evolutionary methods have been applied to several real datasets, with apparent success.