The derivation of quantitative elemental concentrations from multispectral imaging of the Moon has long been a goal of lunar remote sensing. Concentration maps at the spatial resolutions available from the recent Clementine mission would provide a revolutionary new tool for understanding the origin and evolution of the lunar crust. Lucey et al. <1995> presented a method for extracting the concentration of Fe from multispectral imaging of the Moon. This paper examines and quantifies important aspects of that technique left unexamined by Lucey et al. which had the potential to severely limit its utility. These aspects include the effects of maturity, grain size, mineralogy, shading due to topography, and the presence of glass. We also present a new algorithm for derivation of TiO2 from multispectral imaging of both mare and highland units. We find that both techniques are only weakly sensitive to maturity and that they have about 1 wt % accuracy based on examination of the spectral properties and compositions of resolved lunar sampling stations presented by Blewett et al. <1997>. We also discuss these findings in the context of two contrasting views of the effect of composition on lunar spectral properties presented by Pieters and coworkers and Hapke and coworkers. We find the view of Hapke and coworkers to be more consistent with our observations. Using a global mosaic of Clementine multispectral data and these element derivation algorithms, we find that the global modal abundance of FeO is 4.5 wt%¿1 wt% and the global modal abundance of TiO2 is 0.45 wt%¿1 wt%. ¿ 1998 American Geophysical Union