With the Dry Weight Rank (DWR) method of 't Mannetje and Haydock [see Journal of British Grassland Society (1963) 18, 268-275] for botanical analysis in pastures, the dry weight proportions of species are estimated from their first, second and third ranks in dry weight in single quadrats. The yield correction of Haydock and Shaw [see Australian Journal of Experimental Agriculture and Animal Husbandry (1975) 15, 663-670] is used additionally to solve the problem of the respective under- and overestimates of the dry weight proportions of high and low yielding species when these grow in patches. In this paper the DWR method is evaluated by means of computer simulation. Main element of the simulation model is a computer sampling program with which a fictitious vegetation can be sampled with a circular quadrat. The output shows that the DWR method works well using relatively small sampling quadrats with, on average, only a few plants per quadrat, irrespective of the horizontal vegetation structure. In vegetations where species grow patchwise, satisfactory results are also obtained using large quadrats with much more plants (i.e. tens) per quadrat. The reason is that in these cases also minor species can compete successfully for first, second and third ranks. However, it appeared that only a certain degree of patchiness is necessary, and with the usually applied quadrat sizes up to 25 dmsuperscript 2, probably in most vegetations this condition is fulfilled. Care should be taken in applying the DWR method for estimating species composition in recently sown grasslands where species usually occur more or less at random. In those cases, in principle a very small sampling quadrat (smaller than 1 dmsuperscript 2) could be used. However, this has practical limitations since the quadrat size should not be too small for realistic yield estimations, needed for the Haydock & Shaw yield correction. The simulations revealed that one condition (i.e., that the sampling quadrat should be at least as large that it usually contains three or more species) is not necessary because of the almost always perfect functioning of the correction for missing ranks. Generally speaking, a sampling quadrat should be chosen not larger than is strictly necessary from the viewpoint of horizontal vegetation structure and from the viewpoint of realistic yield estimations. Multipliers calculated from simulation data could satisfactorily mimic the original multipliers of DWR given by 't Mannetje & Haydock. It is postulated that the DWR method is well suited for studying vegetation changes in old, floristically diverse grasslands with dominant species often in moderate dry weight proportions and species usually growing in patches.