Many natural activities involve "stopping dilemmas": situations that require a repeated decision between investing effort to achieve some valued goal and stopping that effort to try something else. Previous research into these problems highlights two contradicting biases. While one class of studies suggests a tendency to stop too late (e.g., escalation of commitment), another class of studies suggests a tendency to give up too early (e.g., learned helplessness). Our paper clarifies the conditions that trigger these biases by focusing on two factors: the decision mode (ongoing decisions vs. planning in advance) and the probability each search effort will be costly. We find that experience with stopping dilemmas produces a reversed sunk-cost effect: Most participants stop too early when search is frequently costly but stop too late when search is usually rewarding. This effect can be explained by assuming that stopping decisions reflect reliance on small samples of past experiences with similar stopping dilemmas. Comparison of ongoing and planning decisions reveals an interaction: planning in advance increased search when searching was frequently costly, but decreased search when most search efforts were rewarding. This interaction can be explained by assuming a contingent re-evaluation process: Recent losses increase the tendency to reevaluate a plan to continue the search, and recent gains increase the tendency to reevaluate a plan to stop. In addition, we observe a preference for stopping strategies that imply maximal search. We assume this reflects an attempt to explore the full problem space. (PsycInfo Database Record (c) 2022 APA, all rights reserved).