We evaluate the impact of disasters on income mobility by drawing on "natural experiments". While the poor have a much higher probability of remaining poor when entering a crisis compared to normal times, there is also a negative effect in the year after. Richer households seem to be unaffected. A simple bootstrap method is proposed to facilitate statistical inference for mobility matrices. Also, we simulate measurement error to illustrate its magnitude on these matrices. Small errors induce a substantial downward bias of the probability of remaining poor, while comp arisons across states seem more robust, which is promising for impact analysis.