The authors show that a regularly used temperature extreme metric leads to a systematic underestimation of the expected extreme frequency of up to − 75% and propagates to other derived metrics. A simple bias correction is presented to eliminate this error.