Abstract The short duration of high‐quality, reliable tropical cyclone (TC) observations limits our ability to detect long‐term trends. Here, we analyze TC frequency trends in the NOAA 20th Century Reanalysis (20CR) versions 3 and 2c over the 19th and 20th centuries. TC counts in these data sets are particularly sensitive to TC lifetime maximum wind intensity thresholds, even with the sign of the trends reversed. Higher thresholds identify more organized vortex circulations and reveal observational inconsistencies inherited from assimilated sea‐level‐pressure data. Similar analysis is performed in other reanalysis data sets and tracking algorithms to assess the robustness of our results. Changes in large‐scale environmental fields influencing TC activity since the pre‐industrial era differ markedly between 20CR versions 3 and 2c, further reducing confidence in reanalysis‐based trend assessments. Our results underscore the continuing challenge of inferring robust, long‐term TC frequency changes from historical reanalyses.