We study the IR-through-UV wavelength dependence of 328 Galactic interstellar extinction curves affecting normal, near-main-sequence B and late O stars. We derive the curves using a new technique that employs stellar atmosphere models in lieu of unreddened "standard" stars. Under ideal conditions, this technique is capable of virtually eliminating spectral mismatch errors in the curves. In general, it lends itself to a quantitative assessment of the errors and enables a rigorous testing of the significance of relationships between various curve parameters, regardless of whether their uncertainties are correlated. Analysis of the curves gives the following results:

1. In accord with our previous findings, the central position of the 2175 Å extinction bump is mildly variable, its width is highly variable, and the two variations are unrelated. 2. Strong correlations are found among some extinction properties within the UV region, and within the IR region. 3. With the exception of a few curves with extreme (i.e., large) values of R(V), the UV and IR portions of Galactic extinction curves are not correlated with each other. 4. The large sight line-to-sight line variation seen in our sample implies that any average Galactic extinction curve will always reflect the biases of its parent sample.

The use of an average curve to deredden a spectral energy distribution (SED) will result in significant errors, and a realistic error budget for the dereddened SED must include the observed variance of Galactic curves.

While the observed large sight line-to-sight line variations, and the lack of correlation among the various features of the curves, make it difficult to meaningfully characterize average extinction properties, they demonstrate that extinction curves respond sensitively to local conditions. Thus, each curve contains potentially unique information about the grains along its sight line.