Abstract Crude protein in corn and soybean meal have been documented to vary, and such inherent variability can result in under- or over-feeding of CP when feeds are formulated, leading to reduced bird growth, added input costs, and increased environmental pollution. The purpose of this study was to compare 2 grain-handling techniques and 2 feed formulation methods (linear vs. stochastic programming) to reduce CP variability in finished feeds and determine resulting costs or savings. The 2 grain-handling techniques were placing all the random batches of each delivered ingredient in to (1) a single bin (1-bin method) or (2) segregating above- and below-average samples into 2 bins (2-bin method). A fast way of estimating the composition of the ingredients is now available (near-infrared reflectance spectroscopy). Microsoft Excel workbooks were constructed to solve broiler starter feed formulation problems. Formulating feeds by linear and stochastic models based on the 2-bin method reduced CP variability by at least 50% compared with the 1-bin method. Formula cost was reduced by ˜20 cents per ton (averages of August 2012 United States ingredient prices) when the 2-bin method was used with the linear model. Formulating feed with a margin of safety increased formula cost by $3.40 per ton. Stochastic feed formulation increased formula cost to meet the specified CP level in feed at any probability of success, and formula cost was reduced substantially with the 2-bin method (up to $6.47 per ton). The magnitude of savings and reduced feed variability suggested that, regardless of the costs associated with building extra bins, the 2-bin method can be economically efficient in the long run. Therefore, it could be possible to split the batches of feed ingredients at a feed mill into above- or below-average bins before feed formulation to reduce CP variability and to maximize savings.