Production systems in low-income developing countries are generally poorly diversified, focusing on rainfed staple crop production and raising livestock. These activities are inherently risky and investment and production decisions by farm households are therefore made within environments that are affected by risk. Because of poorly developed or absent credit and insurance markets it is difficult to pass any of these risks to a third party. As a result, it is often found that even when the expected net return is high, households are reluctant to adopt new agricultural technologies when they involve risk. Better understanding risk behavior will be essential for identifying appropriate farm-level strategies for adaptation to climate change by low-income farmers. Despite risk’s potentially central role in farm investment decisions, there have been few attempts to estimate the magnitude and nature of risk aversion of farm households in low-income developing countries. To partially close this gap, this paper uses an experimental approach applied to 262 households in the Ethiopian highlands with real payoffs. By incorporating both small and large stakes and gains and losses into the experiment, we test for the presence of low stake risk aversion and loss aversion. We find that more than 50 percent of the households are severely or extremely risk averse. This contrasts with studies in Asia where most household decision-makers exhibit moderate to intermediate risk aversion. We find that households that stand to lose as well as gain something from participation in games are significantly more risk averse than households playing gains-only games. This strongly suggests that agricultural extension efforts involving losses as well as gains may face systematic resistance by farmers in low-income, high-risk environments. Promotion of technologies with downside risks – even if the upside potential is enormous – should therefore be combined with insurance or other support. We also find that even without the possibility of losses households are much more averse to risk when stakes are high. Results indicate that insurance or other support can likely be phased out. After initial successes have convinced farmers that technologies are viable, risk aversion declines. There are also significant differences in risk averting behavior between relatively poorer and wealthier farm households, which is consistent with decreasing absolute risk aversion. This suggests that as wealth is built up households are willing to take on more risk in exchange for higher returns. Both these findings suggest a strong path dependence. Efforts to develop poor rural areas through promotion of risky technologies should take this path dependence into account. Early successes are important, but households should also be allowed to build up wealth before they are challenged or tempted to take on more risky ventures. Furthermore, the finding that even without the possibility of losses households are much more risk averse when stakes are higher, suggests that agricultural extension should start modestly before asking households to take on larger gambles.