The new theories of endogenous war termination generally predict that initiators would tend to do badly the longer the war, that information acquired during the war would outweigh information available prior to its outbreak, that stronger initiators would be slower to update their estimates about the outcome, and that uncertainty would increase the expected duration of conflict. This article subjects these hypotheses to statistical testing by estimating time-accelerated log-logistic hazard models of duration and bootstrapped ordered probit models of outcome with a new data set of 104 interstate wars from 1816 to 1991. The Monte Carlo simulation results support the hypotheses and the substantive findings provide ample reason for continuing with this research agenda.