Empirical regression models were developed to predict long‐term yields of Hard Red Spring (HRS) wheat (Triticum aestivum L.) and sunflower (Helianthus annuusL.) for soil taxonomic units in North Dakota. Models were developed from onsite soil, climate, management and yield data collected from fertility research plots. In selecting model variables, emphasis was placed on readily available or easily measured agronomic data. Comprehensive models explained 62 and 60% of the variation in wheat and sunflower yield, respectively. Plant‐available water at seeding was the most important variable in wheat and sunflower production. Other variables useful in explaining crop yield variation were those related to soil water‐holding capacity, organic C content, moisture stress, and N availability. Predicted wheat yields compared favorably with current yield estimates for extensive agricultural soils in North Dakota, but predicted sunflower yields were higher than current estimates. The models illustrate a procedure for quantifying soil productivity indexes used for soil interpretation.