The specification of temperature and water vapor profile retrieval accuracy requirements with respect to their bias, rms error and yield relative to truth data, such as 1 K in 1 km thick layers global or 20% in 2 km layers compared to RAOBs, is firmly entrenched as the standard for the evaluation of the performance of hyperspectral sounders. With the availability of forecasts which approach or exceed the 1 K/ 1km accuracy, at least over ocean, and the increasing use of the forecast for the initialization of retrievals, an accuracy metric alone tends to produce misleading results in the evaluation of a retrieval algorithm quality. A much more relevant characterization is retrieval skill, which we define as the ability of an algorithm to get close to the truth, when the truth differs significantly from the forecast, i.e. the anomaly correlation between the retrieved and forecast versus truth and forecast. We explore retrieval skill as function of altitude with a number of regression retrievals, neural net retrievals and physical retrievals of temperature and water vapor profiles over land and ocean using Atmospheric Infrared Sounder (AIRS) and Infrared Atmospheric Sounding Interferometer (IASI) infrared hyperspectral sounder data.