•Soybeans obtained from many 2013-2016 production locations around the United States.•Analysis of total isoflavone and total saponin quantification performed by HPLC.•Preprocessing algorithms were applied to NIRS spectra to minimize variation.•Multiple Linear Regression based models predict isoflavone content.•Predictions had high regression coefficients, low standard errors of calibration. Over 3200 discrete soybean samples were obtained from production locations around the United States during the years 2012–2016. Ground samples were scanned on near infrared spectrometers (NIRS) and analyzed by HPLC for total isoflavone and total saponin composition, as well as total carbohydrate composition. Multiple Linear Regression (MLR) analysis of preprocessed spectral data was used to develop optimized models to predict isoflavone content. The selection of a suitable calibration model was based on a high regression coefficient (R2), and lower standard error of calibration (SEC) values. Robust validated predictions were obtained for isoflavones, however less than robust calibrations were obtained for the total saponins. The correlations were not as robust for predicting the carbohydrate composition. NIRS is a suitable, rapid, nondestructive method to determine isoflavone composition in ground soybeans. Useful isoflavone composition predictions for large numbers of soybean samples can be obtained from quickly obtained NIRS scans.