The results of a 1-yr. experiment in numerical prediction of monthly mean temperatures using a time-averaged thermodynamic model are presented. For the 12-mo. period from December 1965 to November 1966 the statistical evaluation of the prediction of anomalies of the mean monthly surface temperature over North America shows skill higher than persistence for each season, except for summer. Similarly, the evaluation of the predictions of month-to-month changes in ocean surface temperature anomalies also shows skill substantially higher than that of a prediction based on the tendency to return to normal.