An approach to test climate models with observations is presented. In this approach, it is possible to directly observe the longwave feedbacks of the climate system in time series of annual average outgoing longwave spectra. Tropospheric temperature, stratospheric temperature, water vapor, and carbon dioxide have clear and distinctive signatures in the infrared spectrum, and it is possible to detect trends of these signals unambiguously from trends in the outgoing longwave spectrum by optimal detection techniques. This approach is applied to clear-sky data in the tropics simulated from the output of an ensemble of climate models. Estimates of the water vapor–longwave feedback by this approach agree to within estimated errors with truth, and it is likely that an uncertainty of 50% can be obtained in 20 yr of a continuous time series. The correlation of tropospheric temperature and water vapor anomalies can provide a constraint on the water vapor–longwave feedback to 5% uncertainty in 20 yr, or 7% in 10 yr. Thus, it should be possible to place a strong constraint on climate models, which currently show a range of 30% in the water vapor– longwave feedback, in just 10 yr. These results may not hold in the presence of clouds, however, and so it may be necessary to supplement time series of outgoing longwave spectra with GPS radio occultation data, which are insensitive to clouds.