The analysis error of the 500 mb height is estimated as a function of time. The parameters are the observation error, the forecasting error and the network of radiosonde stations from which the analysis centre received observations. With certain assumptions, the average analysis error north of 20° N was found to be 45 m, 22 m, 18 m and 18 m for certain periods in 1946, 1952, 1957 and 1969, respectively. The corresponding maximum errors are 108 m, 76 m, 73 m and 76 m. Areas with possible systematic errors introduced by forecasts and wind observations are determined.