A social group of five saddle-back tamarins (Saguinus fuscicollis) were allowed 15 min per day in a sizeable room adjacent to their home cage. Every other day two additional novel test objects were placed in the room; one contained food on first presentation, and the next day the locations of both were sometimes moved. From the outset, and even when there were 30 objects to choose from, the animals were acute in detecting the novel objects and in remembering the objects and the locations in which they had found food. Whichever individuals had eaten first were among the first to approach the next day. Subsequent tests showed that such one-trial learning was not dependent on object-novelty; that the animals probably remembered all 30 objects and the location of each; and that they spontaneously performed what amounts to generalized delayed matching to sample. The data match or surpass the asymptotic performances of other marmosets on, for example, learning set tasks but are consistent with what is known about the foraging habits of wild S. fuscicollis. Optimal foraging theory is less likely to be an overestimate of animals' mental capacities than previous studies are an underestimate.