AbstractMonthly and annual temperature and precipitation data were studied to determine the most efficient normal for predictive purposes using root mean square error. The number of years over which the normal was calculated was varied, and the value where the minimum error occurred was found to differ considerably from month to month, but there existed some conformity between stations within the same month.A Monte Carlo simulation was developed for the above procedure. The simulation of the station relocation for temperature data at Flemington, New Jersey suggested that the relocation could limit the most efficient length of the climatic normal to a shorter period. The results from a simulation of annual temperature trends at New Brunswick, New Jersey closely agreed with those found with actual data. Thus, the low value found for the most efficient normal is largely due to these trends. Relocations were found to have little effect on the most efficient normal for precipitation data.A modified procedure, varying the number of iterations, showed that RMSE was also dependent on the period of record selected. With this procedure the most efficient normal was found to be between 35 and 40 years for a station with minor relocations, and longer for a benchmark station. In most cases, however, there was little difference between a 10‐year normal and the most efficient norma