2 probability learning experiments evaluated S's ability to adopt optimal strategies of predicting the mode, the median, and the mean. Manipulation of the 3 optimal strategies was accomplished by using different cost schedules in conjunction with a highly skewed probability distribution on a ratio-scaled prediction variable. Ss adopted optimal strategies with respect to the mode and median cost schedules, but their predictions associated with the mean cost schedule deviated in the direction of the median. Exp. II demonstrated that predictions of the mean could be improved by emphasizing the differences between costs proportional to absolute error versus squared error. Implications were discussed for the use of least-square solutions in the simulation of man as an intuitive statistician. (PsycINFO Database Record (c) 2006 APA, all rights reserved)