The probability density function of one-dimensional measurements of ice draft is shown to be fitted accurately by a set of lognormal distributions. A technique is also described for generating synthetic ice draft data by interspersing filtered random number sequences drawn from the fitted lognormal distributions. The sequences are interspersed by random selection to produce one- and two-dimensional model ice draft profiles. The filter characteristics can be chosen so that one-dimensional spectra of the model profiles agree well with measurements. For features with horizontal length scales greater than ~10 m, realizations of ice draft data are in good visual agreement with measured profiles. ¿ American Geophysical Union 1991