Substrate utilization tests are increasingly used for characterizing microbial communities. The applied statistical methods like principal component analysis or detrended correspondence analysis are one-sample methods unsuited for specifying the substrates contributing for separation among groups. Further, these methods demand a high number of replications, a prerequisite that is usually not met. In this paper, a method is proposed that reduces the amount of replicates needed but still allows statistically sound data evaluation. In a first step, in a screening assay with a high number of substrates (31) in three replicates, those substrates are identified that most likely discriminate among the sample types under investigation. In a second step, multivariate analysis of variance and tests based on simultaneous confidence intervals are applied in an assay using this smaller set of substrates (8), but in sixteen replicates. Our approach emphasizes the need of a high ratio of numbers of replicates to the numbers of variables. The substrates contributing most to the separation among groups are determined with a multivariate separation measure, taking the combined effect of several substrates into account. The Mahalanobis distance is calculated to measure distances between the various sample types. The advantage of the approach is that it allows more advanced statistical techniques, like factor analysis and canonical correlation analysis to reduce the variables of different substrate groups, followed by resampling techniques like jackknife and bootstrap algorithms (calculated with Monte Carlo approximation) and Bayes statistics to improve statistical inferences. The approach was tested with a set of three sample types (compost, pasture soil and a mixture of both) and proved suitable for this application.