Models of the geomagnetic main field of degree and order 13 are constructed for epoch 1992.5. These are based on combinations of (quiet-time, low activity) Polar Orbiting Geomagnetic Survey (POGS) total intensity data, and various near-surface, largely vector, measurements. We consider the significance in the fit of models to data of: (a) the spatial sampling interval; (b) the estimated ionospheric field corrections at POGS satellite altitude produced by Quinn et al. (1995); (c) the degree and order one external field coefficients; and (d) the relative proportions of POGS and near-surface data. At this degree and order it is found that: (a) the main-field model accuracy is relatively insensitive to the average sampling interval over a 15:1 range of intervals; (b) the most accurate models are those in which the estimated ionospheric field corrections are not included; (c) the external dipole term is of the order of +20 nT and hence has a marginal impact on model accuracy at the Earth's surface; and (d) a relatively high ratio, of order 4: 1, of POGS to surface data optimizes model accuracy at both POGS satellite altitude and ground level, although only a low ratio of POGS to surface data, of order 1:9, is required for high model accuracy at the Earth's surface alone. Overall, the most accurate main-field model produces spot values at the surface of the Earth with residual standard deviations between 96 nT and 120 nT, depending on component. An estimate of a typical "true", or absolute, model accuracy is also given, by comparing computed values, from a simplified model, with measured data from an independent near-surface data set. It is concluded that the absolute model accuracy at the Earth's surface is of the order of twice the accuracy of the model fit to its input data, again depending on component.