Empirical data-based models of the magnetospheric magnetic field have been widely used during recent years. However, the existing models (Tsyganenko, 1987, 1989a) have three serious deficiencies: (1) an unstable ''de facto'' magnetopause, (2) a crude parametrization by the K-p index, and (3) inaccuracies in the equatorial magnetotail B-z values. This paper describes a new approach to the problem; the essential new features are (1) a realistic shape and size of the magnetopause, based on fits to a large number of observed crossings (allowing a parametrization by the solar wind pressure), (2) fully controlled shielding of tile magnetic field produced by all magnetospheric current systems, (3) new flexible representations for the tail and ring currents, and (4) a new ''directional'' criterion for fitting the model field to spacecraft data, providing improved accuracy for field line mapping. Results are presented from initial efforts to create models assembled from these modules and calibrated against spacecraft data sets.