Numerous studies have revealed the existence of nonrandom trait distribution patterns as a sign of environmental filtering and/or biotic interactions in a community assembly process. A number of metrics with various algorithms have been used to detect these patterns without any clear guidelines. Although some studies have compared their statistical powers, the differences in performance among the metrics under the conditions close to actual studies are not clear. Therefore, the performances of five metrics of convergence and 16 metrics of divergence under alternative conditions were comparatively analyzed using a suite of simulated communities. We focused particularly on the robustness of the performances to conditions that are often uncertain and uncontrollable in actual studies; e.g., atypical trait distribution patterns stemming from the operation of multiple assembly mechanisms, a scaling of trait-function relationships, and a sufficiency of analyzed traits. Most tested metrics, for either convergence or divergence, had sufficient statistical power to distinguish nonrandom trait distribution patterns without uncertainty. However, the performances of the metrics were considerably influenced by both atypical trait distribution patterns and other uncertainties. Influences from these uncertainties varied among the metrics of different algorithms and their performances were often complementary. Therefore, under the uncertainties of an assembly process, the selection of appropriate metrics and the combined use of complementary metrics are critically important to reliably distinguish nonrandom patterns in a trait distribution. We provide a tentative list of recommended metrics for future studies.