A multiple-cone sample splitter, consisting of a series of powder funnels and inverted brass cones mounted alternately in a vertical column over a tray containing small sector-shaped pans,shortens the time necessary to reduce samples to grain-counting size by about 75 percent. A sample design has been set up by which three methods of sample splitting for grain counting--the microsplit, the cone splitter, and hand-quartering--may be compared. The methods of sample splitting are used for one classification in the design and the sample weights of 5, 10, and 20 grams of known grain composition for the other classification. Only one subset of data--that of the 5-gram sample split by the cone splitter--showed a value of chi 2 for precision that was equal to or less than that expected from chance alone. Similar calculations for estimates of accuracy show that the 5-gram sample by the cone splitter and the 10-gram sample by hand-quartering do not exceed the chi 2 values due to chance alone. If the subtotals of the 3x3 experimental design are used to calculate chi 2 , it is seen that: (1) the splitting of the 5-gram sample by all methods is both more accurate and more precise than the splitting of the other two sample weights; and using the three weights for each method (2) the microsplit is the least accurate and the least precise of the three methods, (3) the hand-quartering is slightly more precise than the cone splitter, and (4) the cone splitter is slightly more accurate than hand quartering.