An analysis of the taxonomic composition of 6,496 samples of plants collected for the National Cancer Institute is conducted and the results are presented. The results show that collecting has been highly biased towards woody plant families, as these species are often abundant, they yield samples of multiple parts, and it is often easy to collect sufficient material for isolation of chemicals present in low concentration. Conversely, herbaceous plants that are small or grow in sparse populations are underrepresented in collections. Three of the ten generically most diverse families in Madagascar have not been sampled, and of these ten families, only Rubiaceae and Euphorbiaceae have had greater numbers of samples collected than predicted if collecting was random. The causes of this bias are discussed, and solutions to ensure that bioprospecting efforts gather samples reflective of the diversity present are presented.