Traditional paleontological diversity curves are based on tallies of all taxa appearing in formally defined time units. These tallies are thought to be robust to further data collection. Thus, they supposedly do not reflect nuisance factors like variable time unit lengths and sampling intensity biases. A comparison of a decade-old North American Cenozoic mammal diversity curve and a newer database shows major differences. At least three major factors differentiate the two: use of shorter, fixed-length time intervals; restriction of counts to taxa that cross boundaries between intervals; and correction for variation in sampling intensity. The difference between genus- and species-level data also was examined, but appears to be minor by comparison. Because at best only one pattern can be close to the true historical trajectory, the analyses suggest that the new data and protocols together have yielded a curve that is converging on an accurate signal.