Tritium breeding calculations for a Li17Pb83 benchmark problem that employs steel as structure are presented. Large deviations between the results of continuous energy Monte Carlo and multigroup discrete ordinates are observed when different multigroup libraries are used. Effects of group structure and weighting spectra are explored by collapsing the Los Alamos National Laboratory 80-group library into different broad group structures using different weighting spectra. For blanket systems with natural lithium-lead, many groups with fine structure in the iron resonance region are required for accurate tritium breeding determination. Fewer broad groups can be used only if an appropriate weighting spectrum representing the spectrum in the Li17Pb83 system is used to generate the data. For systems highly enriched in 6Li, these effects are less pronounced with fewer groups being adequate.