Differences in fiber digestibility among high-yielding corn silage hybrids usually are small. Therefore, screening tools should be precise. This study was designed to evaluate the relative influence of several external factors on variability among in situ NDF digestibility values, and estimate the potential precision improvements from increasing the number of steers and of replicate samples per steer. Samples from 3 corn hybrids were used as check hybrids during 85 routine in situ trials conducted over several years. Dry ground whole plant from the check hybrids were weighed into sealed Dacron in situ bags; duplicate bags containing each check hybrid were placed into each of 4 larger mesh bags with other samples being assayed. Within each 48 h run, 4 steers were used. Initially data were averaged by hybrid, run, steer, and mesh bag. These then were analyzed by hybrid using a model that included random effects of run, steer, run × steer interaction, and an error term. Data pooled across hybrids were analyzed using models that included the above effects as well as hybrid. In separate models, the hybrid effect was considered fixed and random. Mean in situ fiber digestibility for check hybrids ranged from 56% to 69%; ranking of these hybrids was consistent across runs. Results from the pooled analysis showed that of the total variability, 56% could be attributed to the error term. Contributions of run, steer, and their interaction were 15%, 20%, and 9%, respectively with contributions being similar when analyzed by hybrid. Repeatability SD of hybrid measurements based on 4 data points per steer was 2.2%. Reproducibility SD of measurements based on different steers and runs was 4.4%. Intra-class correlation of measurements from a hybrid evaluated within the same run using different steers was 0.64. Current results showed that SEM would be reduced by 30 to 40% by increasing the number of steers per run from 1 to 4. Based on 4 steers per run, SEM could be reduced an additional 16% by increasing the number of mesh bag per steer from 1 to 4. Additional improvements in precision from resource re-allocation were minimal.