When an agent fails to make an object function properly, there are two possibilities: the agent did something wrong or something is wrong with the object. As in all problems of confounding, these hypotheses can be disambiguated by varying one factor and holding the other constant: in this case, either by holding the object constant and varying the agent (e.g., by asking for help from others) or by holding the agent constant and varying the object (e.g., by trying another object). Here we show that 16-month-old infants engage in distinct patterns of behavior depending on the relative probability of the competing hypotheses: they ask for help more often when they (rather than the object) are the probable cause of failure; they reach for a new object more often when the object (rather than themselves) is the probable cause of failure.