The fossil record provides direct evidence of how diversity has changed over time, but cannot be taken at face value. Diversity curves constructed from counting taxa in the rock record are seriously biased by unevenness of geographical and stratigraphical sampling effort, inequality in the rock record available for sampling, and inconsistent taxonomic practice. Sample standardization removes some bias, but does not overcome more general incompleteness problems. Modelling that accounts for potential biases is a newer approach but needs accurate estimates of rock record and consistent taxonomic data. Uncertainty remains over whether the steep rise in diversity over the last 100 Ma is real or reflects sampling bias. The repeated rise and fall of marine diversity over time correlates closely with the areal extent of sedimentary deposits and independent estimates of the quality of the fossil record, implying a common driving factor, namely tectonically mediated sea-level change. However, whether changes in diversity are primarily biological in origin, or reflect sampling artefact, remains contentious. There is a distinct possibility that many of the apparent rises and falls in diversity over the Phanerozoic, including most of the ‘mass extinctions’, arise either partially or entirely from rock record bias.