To look successively at sites where several spots of light have appeared in the dark, we cannot simply rely on the image left by these targets on our retina. Our brain has to update target coordinates by taking into account each gaze movement that has taken place. A particular type of brain cell--the quasi-visual (QV) neuron--is assumed to play an important role in this updating by combining target coordinates and eye displacement signals. However, what is exactly this role? Is a QV neuron an element of a working memory that encodes the location of a potential target, or is it pointing to the location of the single goal selected for the next saccade? The two roles theoretically correspond to successive stages of processing: the locations of the optional targets would be stored at one stage, whereas the location of the next selected target would be stored at the subsequent stage. With a task that imposes a choice of goals--the triple-step paradigm--we found evidence that several groups of QV neurons can become simultaneously activated in the monkey's frontal eye field (FEF), suggesting that each group represents a different target location. This supports the hypothesis that the FEF itself contains the spatial information about not yet selected targets.