Sheet and rill erosion has been identified as a major problem in much of the United States. The effect of soil erosion on productivity is the major concern. Off-site damage costs from soil erosion are becoming an increasing concern. This study evaluates soil conservation decisions in a farm-level framework, taking into account all relevant production constraints;The analysis uses two different model formulations to examine the problem. The first is a multiperiod linear programming model which is capable of accounting for the user cost of soil erosion. All production constraints necessary to analyze soil conservation at the farm level are included in the model. The second model is a stochastic dynamic programming model. This model is designed to determine the importance of stochastic soil erosion on the conservation decision. The model also includes tillage machine replacement decisions;The linear programming model results indicate that some conservation should be practiced on all soils in Northeastern Iowa. The major conservation practice is that of minimum tillage. The erosion productivity impact is the most important variable in determining the level of soil conservation. Factors that will impact on soil conservation are the discount rate, tillage machine type on the farm, technology, output prices, farm debt, and an off-site damage tax. The model also supplied some insights into modelling fields versus soils. Field and soil modelling can give different results if the soils in the field have very different characteristics;The dynamic programming model results indicate that the age and type of tillage machines on the farm will have a major effect on the optimum decision. Minimum tillage will be practiced on most fields, but if conventional tillage machines are on the farm, conventional tillage is likely to be practiced. Conventional tillage machines are replaced by minimum tillage machines when they are worn-out. The stochastic soil erosion in the model has no impact on the solution. Stochastic soil erosion might be important if a different objective function, such as maximizing utility, is used in the analysis.