In phenotype annotations curated from the biological and medical literature, considerable human effort must be invested to select ontological classes that capture the expressivity of the original natural language descriptions, and finer annotation granularity can also entail higher computational costs for particular reasoning tasks. Do coarse annotations suffice for certain applications? Here, we measure how annotation granularity affects the statistical behavior of semantic similarity metrics. We use a randomized dataset of phenotype profiles drawn from 57,051 taxon-phenotype annotations in the Phenoscape Knowledgebase. We compared query profiles having variable proportions of matching phenotypes to subject database profiles using both pairwise and groupwise Jaccard (edge-based) and Resnik (node-based) semantic similarity metrics, and compared statistical performance for three different levels of annotation granularity: entities alone, entities plus attributes, and entities plus qualities (with implicit attributes). All four metrics examined showed more extreme values than expected by chance when approximately half the annotations matched between the query and subject profiles, with a more sudden decline for pairwise statistics and a more gradual one for the groupwise statistics. Annotation granularity had a negligible effect on the position of the threshold at which matches could be discriminated from noise. These results suggest that coarse annotations of phenotypes, at the level of entities with or without attributes, may be sufficient to identify phenotype profiles with statistically significant semantic similarity.