In this paper we are investigating how production risk may influence the way a risk averse producer like a subsistence farmer chooses optimal input levels. Risk averse producers will take into account both the mean and the variance of output, and therefore we expect them to choose input levels which differ form the optimal input level of risk neutral producers. Production risk is of particular importance in developing countries, since variance in production here may have grave consequences for the farmer and his family. To model the production decision problem under such circumstances we have made use of the fact that production risk can be treated as heteroskedasticity. Our analysis is based on a dataset obtained from a survey on smallholders in the Kilimanjaro region in Tanzania. Since evidence of output risk in inputs is found, we reestimate the mean and variance function using a maximum likelihood estimator, and correct the standard errors to provide valid inference.