Databases of pollen and spore occurrences have been used to quantify the radiation of angiosperms during the Cretaceous. Although biases are acknowledged, consistency between pollen/spore and leaf records has been used to argue for the reliability of the observed patterns. We present analyses of a high diversity palynoflora (Allon, ca. 125 plant species) from the Santonian of Georgia, USA that illustrate the impact of several of those biases.

Past standard sampling (without replication) and processing (seiving to clean samples) methods markedly reduce apparent diversity. However, the largest impact on diversity occurs when sampling is restricted to only few hundred specimens (usually 300). Not only does restricted sampling fail to capture rare taxa as expected, but also repeatedly counting only 300 specimens even in replicates adds relatively few new rare taxa with each new count. However, expanded sampling (>1000 grains) will recover rare taxa. And many of these rare taxa are likely to have been produced by insect-pollinated plants. This is corroborated by studies of mesofossils containing in situ pollen identical to dispersed pollen types.

These methods reduce observed diversity, explaining in part why the mean richness of published palynological samples is 31 taxa even in the Late Cretaceous. Moreover, these biases may not be distributed evenly through the Cretaceous or among clades. Many early angiosperms were insect-pollinated and thus produced less, often smaller, pollen than wind-pollinated taxa that evolved later. Thus, our record of the Cretaceous radiation may be largely a record of the radiation of wind-pollinated clades that produced large pollen in large amounts. If this is correct, then why does the leaf record show a similar pattern? Preliminary data from extant floras suggest a partial explanation: derived angiosperms tend to retain their leaves for shorter periods of time, and thus shed more leaves on average than basal angiosperms (e.g., magnoliids).

Recent palynologists rarely continue these methods, but publications over the last five decades, which make up the most part of databases, contain data that has passed through these filters and analytical techniques, or reinvestigation, must begin to correct for these biases.