In this research an algorithm is derived for stratifying skewed populations which is much simpler to implement than any of those currently available. It is based on the suggestion by numerous researchers in the field that it is desirable when stratifying skewed populations to arrange for equal coefficients of variation in each subinterval. Our new algorithm makes the breaks in geometric progression and achieves near-equal stratum coefficients of variation when the populations are skewed. Simulation studies on real skewed populations have shown that the new method compares favourably to those commonly used in terms of precision of the estimator of the mean. We also apply the geometric method to the Lavallée-Hidiroglou (1988) algorithm, an iterative method designed specifically for skewed populations. We show that by taking geometric boundaries as the starting points results in most cases in quicker convergence of the algorithm and achieves smaller sample sizes than the default starting points for the same precision. Finally, geometric stratification is applied to the Pareto distribution, a typical model of skewed data. We show that if any finite range of this distribution is broken into a given number of strata, with boundaries obtained using geometric progression, then the stratum coefficients of variation are equal.