Despite its ubiquity in the natural world, polymorphism is commonly disregarded or poorly sampled in phylogenetic analyses due to deliberate sampling strategy, inadequate sampling effort and limited specimen availability. Poor sampling of intraspecific variation engenders differential sampling of morphs within polymorphic species, which could generate conflicting tree topologies by altering the character-based affinity among taxa. To assess the potential magnitude of this impact, Polymorphic Entry Replacement Data Analysis (PERDA) was developed as a new script for the TNT phylogenetic program. This script simulates poor sampling of polymorphic taxa on a matrix of discrete characters by iteratively replacing each polymorphic state (e.g. [01]) with a randomly selected single state included in the original polymorphic coding (e.g. 0 or 1). The trees recovered from these subsampled data sets provide a distribution of tree distances, which indicates the level of incongruent trees resulting from different combinations of single states. Performing PERDA on empirical data sets shows alarming frequencies and magnitudes of conflicting tree topologies, demonstrating that poor sampling within polymorphic taxa could yield highly incompatible trees in many data sets. This troubling outcome undermines phylogenetic inferences based on data with poor intraspecific sampling, which is typical for palaeontological studies. With trees obtained from subsampled data sets, PERDA also generates a metaconsensus tree revealing interspecific relationships that become ambiguous due to documented levels of intraspecific variation. These collapsed clades point to taxa for which evidence should be sought to justify their taxonomic classification.