The theory of optimum interpolation analysis is presented, with emphasis on the role that observational errors have in the analysis scheme. It is shown that an estimate of the root-mean-square observational error is required and also that the correlations between errors of observations, if nonzero, must be specified. Methods of determining these quantities from observational data statistics are discussed and examples shown. Finally, error-checking routines that either accept or reject data are described. It is pointed out that care should be exercised when checking for errors so that good and useful data are not inadvertently rejected.