Abstract: Abstract Recent contest experience can influence an individual’s behaviour in subsequent contests. When the probability of winning a subsequent contest is used to quantify experience effects, a loser effect usually lasts longer than a winner effect. This conclusion, however, may be caused by this probability understating the persistence of the influence of a winning experience on contest decisions. Using Kryptolebias marmoratus, a mangrove killifish, as the study organism, we investigated whether different conclusions about the relative persistence of winning and losing experiences would be reached when different aspects of contest behaviour (probability of initiating attacks, probability of winning non-escalated and escalated contests, escalation rate and contest duration) were measured. The results indicated that the apparent persistence of the effect of winning or losing experiences varied with the behaviour studied. When the likelihood to initiate attacks was used, no winner effect was detected while the loser effect lasted for <1 d. When escalation rate was used, the winner effect lasted for 2–4 d, while the loser effect lasted for 1–2 d. When the probability of winning non-escalated contests was used, the winner effect was detectable for <1 d, while the loser effect lasted for 2–4 d. And, when contest duration was used, the winner effect was detectable for 2–4 d, but no loser effect was detectable. These results show that (1) the probability of winning a subsequent contest understated the persistence of the influence of a winning experience on the fish’s contest decisions, (2) the measures most effective at detecting winner effects are different from those most effective at detecting loser effects and (3) in K. marmoratus, both effects can be detected 2 d after the completion of experience training but both dissipate in 4 d.