Two observed states of the atmosphere that are very close by some measure, also applied to states of a model. Formal measures of closeness include anomaly correlation, root-mean-square distance, and covariance. Usually one expects analogs to occur only during the same time of year. Atmospheric analogs that are close compared to current levels of observational error are unlikely to be found unless one studies a single variable confined to a very small area (≤1000 km radius), or otherwise reduces the degrees of freedom to a very small number (≤3). Analogs have a practical application in specifying the surface weather from a 3D atmospheric state produced by NWP and in short- term climate prediction for forecasts ranging from a month to several seasons. Analogs are also of interest in research related to predictability, short-term forecasts and error growth, cluster analysis, and estimates of dimensionality of the atmospheric attractor. When natural analogs are poor, improvements have been sought by combining several of them, by using anti-analogs (antilogs), and by constructing an artificial close analog by some objective weighted averaging of a set of previously observed states (constructed analogs). See MOS, perfect prognosis method.