Cognitive models of choice almost universally implicate se-quential evidence accumulation as a fundamental element ofthe mechanism by which preferences are formed. When to stop evidence accumulation is an important question that suchmodels do not currently try to answer. We present the first cog-nitive model that accurately predicts stopping decisions in in-dividual economic decisions-from-experience trials, using anonline learning model. Analysis of stopping decisions acrossthree different datasets reveals three useful predictors of sam-pling duration - relative evidence strength, how long it takesparticipants to see all rewards, and a novel indicator of con-vergence of an underlying learning process, which we call pre-dictive volatility. We quantify the relative strengths of thesefactors in predicting observers’ stopping points, finding thatpredictive volatility consistently dominates relative evidencestrength in stopping decisions.