Researchers have shown that statistical learning is a pervasive and deeply entrenched human ability. As the input environment becomes more predictable, our processing of that environment should show increasing structure. In this study, we capture these unfolding processes using the dynamics of arm movements and reveal explicit patterns of predictive arm placement during learning. We tracked the arm using the Nintendo Wii remote while participants engaged in a visuospatial statistical learning task. Over training, the arm exhibits predictive movements towards event sequences that have higher transitional probabilities. With only a short period of training in few subjects, these anticipatory movements reflect the emergence of statistical structure in a visual learning space.