Spherical harmonic models were derived using observatory data taken since 1995 and Ørsted vector data. Such models were used to predict both the field in 2005 and estimated error surfaces. Using only satellite data taken over the 2-year interval produced unsatisfactory predictions of the field. It was essential to include longer period observations at fixed observatories and other surface data, such as ship-towed data. To accurately determine the field for past epochs it is necessary to use parabolic terms in time to degree n = 12 as well as annual terms to allow for external contributions. However, using only linear terms provided the best accuracy for projections to 2005. Reduction of the complexity of the model to n = 8 tends to smooth over areas devoid of a long period of observation though slightly decreasing the accuracy elsewhere. The best estimates for 2005 using these techniques cannot be more accurate than a few hundred nT at the Earth's surface.