A reanalysis of the Experimental Breeder Reactor-I meltdown shows that far better agreement with experiment is obtained by the assumption that the coolant feedback coefficient operating during the later stages of the transient is positive and unstabilizing. A possible reason for such an occurrence is suggested.