Evaluation of claimed “leveling off” of warming in satellite temperature data. Results are for monthly-mean anomalies in the temperature of the mid- to upper troposphere (TMT), corrected for stratospheric cooling8 and spatially averaged over 82.5°N–82.5°S. The average of the latest satellite dataset versions (RSS v4.0, STAR v4.0, and UAH v6.0) has a warming trend of 0.181 °C/decade over the 456-month period from January 1979 to December 2016 (panel A). Maximally overlapping 20-year (240-month) trends in the six individual satellite TMT time series are plotted on the end month of the trend-fitting period (panel B). The p-values for these trends (panel C) are for tests of the null hypothesis that observed tropospheric warming could be due to natural internal variability alone3. The grey shaded box is the rejection region (at a stipulated 10% significance level) for the null hypothesis. The p-value calculations rely on estimates of the multi-decadal internal variability of the climate system from model pre-industrial control runs. These simulations have no year-to-year changes in natural or human external forcings. The sampling distributions of control run TMT trends on 20-year and 38-year timescales (the orange histograms in panels D and E, respectively) are based on results from 36 different models. The symbols (plus signs) in panel D are the final 20-year trends from panel B. Bold vertical lines in panel D are the averages of the overlapping 20-year trends in panel B (see arrows to the right of panel B). Vertical lines in panel E are observed trends over the full 456-month satellite record. Results in D and E are for the latest satellite dataset versions only. Full analysis details are in ref. 3 and the Methods section.