Groups of plant systematists at the Royal Botanic Gardens, Sydney, were given 87 pictures of plant parts in one of two classes, and asked to divide them into character states in two separate experiments. In the first experiment (Trial 1) the pictures were of cotyledons only. In the second experiment (Trial 2), the pictures constituted of triplets: a cotyledon, seedling leaf, and inflorescence bract. These triplets were intended to simulate complex data such a might be garnered by looking at a plant. The pictures were drawn from a published treatment of the plant genus Banksia. Each picture or triplet represented one taxon. In each trial the systematists were given about ¾ hour to create character states (groups of similar pictures). Hierarchical groupings were not permitted. Each experiment resulted in four characters, one for each group of systematists. For analysis, these characters were represented as trees, and consensus trees computed. Visual inspection of the data trees showed that the systematists were able to produce smaller, more precisely defined clusters (character states) in Trial 2. The Strict and Majority Rule consensus trees for Trial 1 were completely unresolved, while those for Trial 2 contained a single cluster of two triplets, and three clusters of two triplets, respectively. The Semi-strict tree for Trial 1 contains four clusters, two of which are nested in a third. The average cluster size is 6.3. The Trial 2 tree contains seven clusters, two of which are nested in a much larger third. Average cluster size is 4.4. The main differences between the two Adams trees are the number and sizes of the clusters. The Trial 2 tree has a greater number of smaller clusters than the Trial 1 tree. Conclusion: Rich data does allow greater inter-investigator agreement in character states.

Key words: Banksia, character states, characters, complexity, morphological characters, phylogenetic analysis