In designing inventories of wildland vegetation, two of the many basic methodological choices are: 1) whether data are collected, reduced, and stored in discrete classes or as continuous variables, and 2) whether data are gathered as general purpose variables to bear upon many questions, or as specific purpose variables optimized for only one type of prediction. The effects of these two choices on accuracy of vegetation inventories to predict plant community production were examined by comparing regression models built upon differing sets of independent variables "inventoried" from a common data base. Contrary to expectations, discrete variables of classified community types were better predictors of plant community production than the same vegetation data reduced as continuous variables by three ordination techniques. Substitution of specific purpose soil and vegetation variables thought to be especially relevant to production did not improve correlations from those of analogous general purpose variables. These results do not show the anticipated accuracy loss of general purpose inventory variables, but such findings cannot yet be generalized to other situations. Implications for the design of practical, extensive survey methods for wildland vegetation are briefly discussed.