A simple model is developed relating the debt and asset portfolio of the farm to the production decision, which leads to a small non-linear system of equations. The system is estimated with time-series cross-sectional data from Australian broadacre agriculture using non-linear three-stage least squares. This gives a new method of estimating risk aversion coefficients by using actual behaviour of farmers in a realistic economic environment, rather than games played in artificial situations. Australian farmers are found to be risk averse, and the partial coefficient of risk aversion decreases with wealth and increases with income. The results are consistent with the results of studies by Binswanger in India and elsewhere using a completely different method. This consistency suggests that the partial risk aversion coefficient is a relatively robust measure of attitudes to risk.