Researchers have proposed that people set thresholds to decide when to stop searching in optimal stopping tasks with full information, where option values are known. Most models assume that individuals set internal thresholds to guide their stopping decisions. However, whether humans actually set and adjust thresholds with experience remains unexamined. This experiment investigates how people set and adjust thresholds and whether this affects search behavior and learning over time. We designed an optimal stopping task where participants either report a threshold before seeing the option’s value or proceed without setting one. In addition, we varied whether the set threshold was binding for stopping decisions. Our findings, based on model predictions and empirical data, suggest that setting thresholds leads to more errors and lower accuracy. Accuracy is lowest when thresholds are non-binding. Participants often deviate from their set thresholds and perform better for doing so. These results challenge the assumption that people rely on thresholds for stopping decisions. Instead, they seem to learn from experience to improve accuracy and reduce errors, offering new insights into sequential decision making.