Standard economic risk theory postulates that in the absence of credit markets, wealthier households will engage in higher-risk, higher profit activities to generate income while poor households will specialize in low-risk activities with low returns. The rationale is that wealthier households can deplete savings when things go wrong whereas poor household cannot. This theoretical argument has been tested for several countries and is generally validated by the data. However, existing studies on the relation between savings and activity choices implicitly assume that savings are certain or risk-free. This study suggests that explicitly allowing household savings or assets to be risky can yield results that differ considerably from the pattern predicted by the standard theoretical model. Using data from the 1998 household priority survey in Burundi, we estimate the relationship between household savings (livestock) and choices of income- generating activities (risky vs. less-risky activities). We exploit the fact that surveyed households in certain regions in Burundi were exposed to a relative higher level of risk and uncertainty due to the civil war preceding and during the time of the survey. We find that in general household savings exercise their usual risk-taking effect, though that this effect disappears and even reverses for households in the conflict affected regions. In those regions, wealthier households do not reduce allocation to low-risk low-return activities. We argue that this finding can probably (in part) explain the massive increase in poverty in the provinces exposed to the war during the 1990-1998 period. In this fashion, we argue that a type of ’productive social safety net’, as recently discussed in the development literature, could possibly be an effective policy measure to lower the increased asset risk induced by conflict.